[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16731614#comment-16731614 ] ASF GitHub Bot commented on OMID-90: Github user yonigottesman closed the pull request at: https://github.com/apache/incubator-omid/pull/46 > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Fix For: 1.0.0 > > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16655320#comment-16655320 ] James Taylor commented on OMID-90: -- Everything is good with OMID-109. All the transaction tests pass. Last run the failure was a known flapper, unrelated to our work. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16655316#comment-16655316 ] Yonatan Gottesman commented on OMID-90: --- [~jamestaylor] Im not commiting yet because stuff looks like they still fail in OMID-109. lets stabilize that first > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16654279#comment-16654279 ] James Taylor commented on OMID-90: -- {quote}Where would one go to edit stuff like the command used to compile and stuff like that? {quote} Good question. Unfortunately I don't know either. [~mujtabachohan] has always graciously made these changes. However, since our builds are working now, can't you just check-in something to Omid's phoenix-integration branch to make low latency the default and then kick off an omid2 Jenkins job? And then just revert it back. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16654145#comment-16654145 ] Yonatan Gottesman commented on OMID-90: --- Hi, [~jamestaylor] I dont have any experience with editing jenkins jobs. Where would one go to edit stuff like the command used to compile and stuff like that? > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16654068#comment-16654068 ] Ohad Shacham commented on OMID-90: -- Awesome, Thx. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16654065#comment-16654065 ] James Taylor commented on OMID-90: -- Yes, Yoni has the permissions he needs now to change the jenkins job. He could potentially change the default temporarily just to get a test run in. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16654048#comment-16654048 ] Ohad Shacham commented on OMID-90: -- I assume that [~yonigo] can commit the low latency, we already reviewed everything. He wanted to wait until everything passes before he does the commit. The low latency is not default, [~jamestaylor] how can we change the configuration when running the jenkins? Do we have permissions to do some? > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16653654#comment-16653654 ] James Taylor commented on OMID-90: -- Ok, sounds like we're good, then. Is it ok if [~yonigo] commits this now, [~ohads]? Would it be possible to get a Phoenix test run on the omid2 branch with the low latency version? > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16653340#comment-16653340 ] Ohad Shacham commented on OMID-90: -- {quote}In PHOENIX-4943, I write the shadow cells of the index rows when the index rows themselves are written. Does that solve the issue? {quote} Yes [~jamestaylor], the fence functionality exists in the transaction manager. The thing is that since during index population Phoenix writes that rows with the fence id, then before the fence id is not written to the commit table, we have to guarantee that the rows are written with the shadow cells. Otherwise, there might be a case that someone will try to read before the shadow cells exists. On a seconds thought, maybe it is not possible since the index is not valid yet? In any case, we want to do it since writing once is better than writing and then writing again the shadow cells. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16652959#comment-16652959 ] Yonatan Gottesman commented on OMID-90: --- Hi I dont want to mixup the low latency changes before all tests pass so we dont have trouble debugging. when phoenix branch in stable and 109 runs without errors i will merge low latency after testing with phoenix. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16652519#comment-16652519 ] James Taylor commented on OMID-90: -- FYI, the test that inserts rows as the index is being initially populated is BaseIndexIT.testCreateIndexAfterUpsertStarted(). That'll be tested when GlobalMutableTxIndexIT is run. Would be good to try this with the low latency version of Omid. Even better would be to try a complete run with the low latency version. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16651850#comment-16651850 ] James Taylor commented on OMID-90: -- In PHOENIX-4943, I write the shadow cells of the index rows when the index rows themselves are written. Does that solve the issue? The fence was more to guard against the case in which a write to the data table occurs before we start maintaining the index. Prior to PHOENIX-4943, we were writing the shadow cells in the standard way (i.e. after the commit succeeded). There's a test for this - would be good if you guys ran the Phoenix unit tests with the low latency version. I don't remember the particular test that tests writing to a table while the index is being created - [~tdsilva] - do you remember? > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16651230#comment-16651230 ] Ohad Shacham commented on OMID-90: -- [~jamestaylor], sorry for the late response. In the low latency mode, the fence information is not written to the commit table and the fence id is only returned to the client. This is why using auto commit in this case is important. In -PHOENIX-4943- you solved this issue, I just wanted to make sure it does. In the non low latency mode the fence info is written to the commit and can hide an incorrect auto commit (with runtime penalty of course :)). > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16650361#comment-16650361 ] James Taylor commented on OMID-90: -- Ping, [~ohads]. I didn't understand your question: {quote}This commit removes the addition of the fence information to the commit table. Could you please check if it hurts correctness? It is related to the index population with auto commit that you have just added. {quote} We need the fence functionality in Phoenix so that when an index is created, we can guarantee that we don't miss creation of any data rows. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16646680#comment-16646680 ] James Taylor commented on OMID-90: -- If you could please commit the smaller OMID-117 before this one to save any merge headaches, that be appreciated. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16646643#comment-16646643 ] James Taylor commented on OMID-90: -- [~ohads] - which commit? I'm not sure what you're asking. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16646350#comment-16646350 ] Ohad Shacham commented on OMID-90: -- [~jamestaylor], This commit removes the addition of the fence information to the commit table. Could you please check if it hurts correctness? It is related to the index population with auto commit that you have just added. Thx! > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16645330#comment-16645330 ] ASF GitHub Bot commented on OMID-90: Github user JamesRTaylor commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r224179421 --- Diff: hbase-client/src/test/java/org/apache/omid/transaction/TestOmidLLRaces.java --- @@ -0,0 +1,243 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.omid.transaction; + + +import static com.google.common.base.Charsets.UTF_8; +import static org.apache.omid.committable.hbase.HBaseCommitTableConfig.DEFAULT_COMMIT_TABLE_CF_NAME; +import static org.mockito.Mockito.doAnswer; +import static org.mockito.Mockito.spy; +import static org.testng.Assert.assertFalse; +import static org.testng.Assert.assertTrue; +import static org.mockito.Matchers.any; + +import com.google.common.base.Optional; +import org.apache.hadoop.hbase.TableName; +import org.apache.hadoop.hbase.client.Get; +import org.apache.hadoop.hbase.client.Put; +import org.apache.hadoop.hbase.client.Result; +import org.apache.hadoop.hbase.client.Table; +import org.apache.hadoop.hbase.util.Bytes; +import org.apache.omid.committable.CommitTable; +import org.apache.omid.committable.hbase.KeyGenerator; +import org.apache.omid.committable.hbase.KeyGeneratorImplementations; +import org.apache.omid.metrics.NullMetricsProvider; +import org.apache.omid.tso.client.OmidClientConfiguration; +import org.apache.omid.tso.client.TSOClient; +import org.apache.omid.tso.client.TSOFuture; +import org.mockito.invocation.InvocationOnMock; +import org.mockito.stubbing.Answer; +import org.testng.ITestContext; +import org.testng.annotations.Test; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.io.IOException; +import java.util.Set; +import java.util.concurrent.CountDownLatch; + + +@Test(groups = "sharedHBase") +public class TestOmidLLRaces extends OmidTestBase { + +private static final byte[] row1 = Bytes.toBytes("test-is-committed1"); +private static final byte[] row2 = Bytes.toBytes("test-is-committed2"); +private static final byte[] family = Bytes.toBytes(TEST_FAMILY); +private static final byte[] qualifier = Bytes.toBytes("testdata"); +private static final byte[] data1 = Bytes.toBytes("testWrite-1"); + +private static final Logger LOG = LoggerFactory.getLogger(TestOmidLLRaces.class); +@Override +protected boolean isLowLatency() { +return true; +} + +@Test(timeOut = 30_000) +public void testIsCommitted(ITestContext context) throws Exception { +AbstractTransactionManager tm = (AbstractTransactionManager)newTransactionManagerHBaseCommitTable(getClient(context)); + +Table htable = connection.getTable(TableName.valueOf(TEST_TABLE)); +SnapshotFilterImpl snapshotFilter = new SnapshotFilterImpl(new HTableAccessWrapper(htable, htable), +tm.getCommitTableClient()); +TTable table = spy(new TTable(htable, snapshotFilter, false)); + +HBaseTransaction t1 = (HBaseTransaction) tm.begin(); + +Put put = new Put(row1); +put.addColumn(family, qualifier, data1); +table.put(t1, put); +tm.commit(t1); + +HBaseTransaction t2 = (HBaseTransaction) tm.begin(); +put = new Put(row2); +put.addColumn(family, qualifier, data1); +table.put(t2, put); +table.flushCommits(); + +HBaseTransaction t3 = (HBaseTransaction) tm.begin(); +put = new Put(row2); +put.addColumn(family, qualifier, data1); +table.put(t3, put); +tm.commit(t3); + +HBaseCellId hBaseCellId1 = new HBaseCellId(table, row1, family, qual
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644682#comment-16644682 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223989230 --- Diff: transaction-client/src/main/java/org/apache/omid/transaction/AbstractTransactionManager.java --- @@ -350,6 +356,43 @@ private void markReadOnlyTransaction(AbstractTransaction readO } +private void commitLowLatencyTransaction(AbstractTransaction tx) +throws RollbackException, TransactionException { +try { + +long commitTs = tsoClient.commit(tx.getStartTimestamp(), tx.getWriteSet(), tx.getConflictFreeWriteSet()).get(); +boolean committed = commitTableWriter.atomicAddCommittedTransaction(tx.getStartTimestamp(),commitTs); +if (!committed) { +// Transaction has been invalidated by other client +rollback(tx); + commitTableClient.completeTransaction(tx.getStartTimestamp()); +rolledbackTxsCounter.inc(); +throw new RollbackException("Transaction " + tx.getTransactionId() + " got invalidated"); +} +certifyCommitForTx(tx, commitTs); +updateShadowCellsAndRemoveCommitTableEntry(tx, postCommitter); + +} catch (ExecutionException e) { +if (e.getCause() instanceof AbortException) { // TSO reports Tx conflicts as AbortExceptions in the future +rollback(tx); +rolledbackTxsCounter.inc(); +throw new RollbackException("Conflicts detected in tx writeset", e.getCause()); +} + +if (e.getCause() instanceof ServiceUnavailableException || e.getCause() instanceof ConnectionException) { +errorTxsCounter.inc(); +rollback(tx); // Rollback proactively cause it's likely that a new TSOServer is now master --- End diff -- the leader > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644680#comment-16644680 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223976065 --- Diff: hbase-client/src/test/java/org/apache/omid/transaction/TestEndToEndScenariosWithHA.java --- @@ -287,7 +287,7 @@ public void testScenario1() throws Exception { // TX 2 modifies cells R1C1 & R2C2 (v2) // TX 2 commits // End of Test state: R1C1 & R2C2 (v2) -@Test(timeOut = 60_000) +@Test(timeOut = 60_00) --- End diff -- Why this is needed? Due to the low_cpu mode? or you just forgot it? :) > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644676#comment-16644676 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223975170 --- Diff: hbase-client/src/main/java/org/apache/omid/transaction/SnapshotFilterImpl.java --- @@ -181,22 +179,44 @@ public CommitTimestamp locateCellCommitTimestamp(long cellStartTimestamp, long e // 2) Then check the commit table // If the data was written at a previous epoch, check whether the transaction was invalidated -Optional commitTimeStamp = commitTableClient.getCommitTimestamp(cellStartTimestamp).get(); -if (commitTimeStamp.isPresent()) { -return commitTimeStamp.get(); +boolean invalidatedByOther = false; +Optional commitTimestampFromCT = commitTableClient.getCommitTimestamp(cellStartTimestamp).get(); +if (commitTimestampFromCT.isPresent()) { +if (isLowLatency && !commitTimestampFromCT.get().isValid()) +invalidatedByOther = true; +else +return commitTimestampFromCT.get(); } // 3) Read from shadow cell -commitTimeStamp = readCommitTimestampFromShadowCell(cellStartTimestamp, locator); +Optional commitTimeStamp = readCommitTimestampFromShadowCell(cellStartTimestamp, locator); if (commitTimeStamp.isPresent()) { return commitTimeStamp.get(); } +// In case of LL, if found invalid ct cell, still must check sc in stage 3 then return +if (invalidatedByOther) { +assert(!commitTimestampFromCT.get().isValid()); +return commitTimestampFromCT.get(); +} + // 4) Check the epoch and invalidate the entry // if the data was written by a transaction from a previous epoch (previous TSO) -if (cellStartTimestamp < epoch) { +if (cellStartTimestamp < epoch || isLowLatency) { boolean invalidated = commitTableClient.tryInvalidateTransaction(cellStartTimestamp).get(); if (invalidated) { // Invalid commit timestamp + +// If we are running lowLatency Omid, we could have manged to invalidate a ct entry, +// but the committing client already wrote to shadow cells: --- End diff -- This can happen only in low latency mode, since in the regular mode the client keeps the commit table entry if persisting the commit was done after the tso lost its lease. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644678#comment-16644678 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223974621 --- Diff: hbase-client/src/main/java/org/apache/omid/transaction/SnapshotFilterImpl.java --- @@ -181,22 +179,44 @@ public CommitTimestamp locateCellCommitTimestamp(long cellStartTimestamp, long e // 2) Then check the commit table // If the data was written at a previous epoch, check whether the transaction was invalidated -Optional commitTimeStamp = commitTableClient.getCommitTimestamp(cellStartTimestamp).get(); -if (commitTimeStamp.isPresent()) { -return commitTimeStamp.get(); +boolean invalidatedByOther = false; +Optional commitTimestampFromCT = commitTableClient.getCommitTimestamp(cellStartTimestamp).get(); +if (commitTimestampFromCT.isPresent()) { +if (isLowLatency && !commitTimestampFromCT.get().isValid()) +invalidatedByOther = true; +else +return commitTimestampFromCT.get(); } // 3) Read from shadow cell -commitTimeStamp = readCommitTimestampFromShadowCell(cellStartTimestamp, locator); +Optional commitTimeStamp = readCommitTimestampFromShadowCell(cellStartTimestamp, locator); if (commitTimeStamp.isPresent()) { return commitTimeStamp.get(); } +// In case of LL, if found invalid ct cell, still must check sc in stage 3 then return +if (invalidatedByOther) { +assert(!commitTimestampFromCT.get().isValid()); +return commitTimestampFromCT.get(); +} + // 4) Check the epoch and invalidate the entry // if the data was written by a transaction from a previous epoch (previous TSO) -if (cellStartTimestamp < epoch) { +if (cellStartTimestamp < epoch || isLowLatency) { boolean invalidated = commitTableClient.tryInvalidateTransaction(cellStartTimestamp).get(); if (invalidated) { // Invalid commit timestamp + +// If we are running lowLatency Omid, we could have manged to invalidate a ct entry, --- End diff -- managed > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644679#comment-16644679 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223973121 --- Diff: hbase-client/src/main/java/org/apache/omid/transaction/SnapshotFilterImpl.java --- @@ -181,22 +179,44 @@ public CommitTimestamp locateCellCommitTimestamp(long cellStartTimestamp, long e // 2) Then check the commit table // If the data was written at a previous epoch, check whether the transaction was invalidated -Optional commitTimeStamp = commitTableClient.getCommitTimestamp(cellStartTimestamp).get(); -if (commitTimeStamp.isPresent()) { -return commitTimeStamp.get(); +boolean invalidatedByOther = false; +Optional commitTimestampFromCT = commitTableClient.getCommitTimestamp(cellStartTimestamp).get(); +if (commitTimestampFromCT.isPresent()) { +if (isLowLatency && !commitTimestampFromCT.get().isValid()) +invalidatedByOther = true; +else +return commitTimestampFromCT.get(); } // 3) Read from shadow cell -commitTimeStamp = readCommitTimestampFromShadowCell(cellStartTimestamp, locator); +Optional commitTimeStamp = readCommitTimestampFromShadowCell(cellStartTimestamp, locator); if (commitTimeStamp.isPresent()) { return commitTimeStamp.get(); } +// In case of LL, if found invalid ct cell, still must check sc in stage 3 then return --- End diff -- Please add this comment: "This check is not needed in a non low latency mode, since during post commit, the client keeps the entry in the commit table when persisting the commit was done after the TSO lost its lease." > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644683#comment-16644683 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223974301 --- Diff: hbase-client/src/main/java/org/apache/omid/transaction/SnapshotFilterImpl.java --- @@ -181,22 +179,44 @@ public CommitTimestamp locateCellCommitTimestamp(long cellStartTimestamp, long e // 2) Then check the commit table // If the data was written at a previous epoch, check whether the transaction was invalidated -Optional commitTimeStamp = commitTableClient.getCommitTimestamp(cellStartTimestamp).get(); -if (commitTimeStamp.isPresent()) { -return commitTimeStamp.get(); +boolean invalidatedByOther = false; +Optional commitTimestampFromCT = commitTableClient.getCommitTimestamp(cellStartTimestamp).get(); +if (commitTimestampFromCT.isPresent()) { +if (isLowLatency && !commitTimestampFromCT.get().isValid()) +invalidatedByOther = true; +else +return commitTimestampFromCT.get(); } // 3) Read from shadow cell -commitTimeStamp = readCommitTimestampFromShadowCell(cellStartTimestamp, locator); +Optional commitTimeStamp = readCommitTimestampFromShadowCell(cellStartTimestamp, locator); if (commitTimeStamp.isPresent()) { return commitTimeStamp.get(); } +// In case of LL, if found invalid ct cell, still must check sc in stage 3 then return +if (invalidatedByOther) { +assert(!commitTimestampFromCT.get().isValid()); +return commitTimestampFromCT.get(); +} + // 4) Check the epoch and invalidate the entry --- End diff -- or invalidate in a non latency mode > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644677#comment-16644677 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223971768 --- Diff: hbase-client/src/main/java/org/apache/omid/transaction/SnapshotFilterImpl.java --- @@ -181,22 +179,44 @@ public CommitTimestamp locateCellCommitTimestamp(long cellStartTimestamp, long e // 2) Then check the commit table // If the data was written at a previous epoch, check whether the transaction was invalidated --- End diff -- This "If the data was written at a previous epoch," can be removed. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644681#comment-16644681 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223975342 --- Diff: hbase-client/src/main/java/org/apache/omid/transaction/SnapshotFilterImpl.java --- @@ -240,7 +261,8 @@ public CommitTimestamp locateCellCommitTimestamp(long cellStartTimestamp, long e CellUtil.cloneQualifier(cell), cell.getTimestamp()), commitCache, -tableAccessWrapper)); +tableAccessWrapper), +isLowLatency); --- End diff -- indent > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644684#comment-16644684 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r223993765 --- Diff: tso-server/src/main/java/org/apache/omid/tso/RequestProcessorSkipCT.java --- @@ -0,0 +1,92 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.omid.tso; + +import com.google.inject.Inject; +import org.apache.omid.metrics.MetricsRegistry; +import org.jboss.netty.channel.Channel; + +import java.io.IOException; + +public class RequestProcessorSkipCT extends AbstractRequestProcessor { + + +private final ReplyProcessor replyProcessor; + +private final LeaseManagement leaseManager; +private final Panicker panicker; +private final String tsoHostAndPort; + +@Inject +RequestProcessorSkipCT(MetricsRegistry metrics, + TimestampOracle timestampOracle, + ReplyProcessor replyProcessor, + Panicker panicker, + LeaseManagement leaseManager, + TSOServerConfig config, + LowWatermarkWriter lowWatermarkWriter, + String tsoHostAndPort) throws IOException { +super(metrics, timestampOracle, panicker, config, lowWatermarkWriter); +this.replyProcessor = replyProcessor; +this.tsoHostAndPort = tsoHostAndPort; +requestRing = disruptor.start(); +this.leaseManager = leaseManager; +this.panicker = panicker; +} + +private void commitSuicideIfNotMaster() { +if (!leaseManager.stillInLeasePeriod()) { +panicker.panic("Replica " + tsoHostAndPort + " lost mastership whilst flushing data. Committing suicide"); +} +} + +@Override +public void forwardCommit(long startTimestamp, long commitTimestamp, Channel c, MonitoringContext monCtx) { +commitSuicideIfNotMaster(); --- End diff -- Add a comment that his is required since returning commit results when the TSO is not the leader might violate snapshot isolation. This is because we have to guarantee that committing transaction from previous tso has to persist all its data because a new transaction started by a new tso. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis an
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644675#comment-16644675 ] ASF GitHub Bot commented on OMID-90: Github user ohadshacham commented on a diff in the pull request: https://github.com/apache/incubator-omid/pull/46#discussion_r22397 --- Diff: hbase-client/src/test/java/org/apache/omid/transaction/OmidTestBase.java --- @@ -89,6 +88,8 @@ public void beforeGroups(ITestContext context) throws Exception { TSOServerConfig tsoConfig = new TSOServerConfig(); tsoConfig.setPort(1234); tsoConfig.setConflictMapSize(1000); +tsoConfig.setLowLatency(isLowLatency()); +tsoConfig.setWaitStrategy("LOW_CPU"); --- End diff -- nice > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644500#comment-16644500 ] Yonatan Gottesman commented on OMID-90: --- [~giacomotaylor] [~ohads] I added a patch to integrate this with phoenix branch. The patch applies on top of OMID-115. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf, omid90.patch > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16644495#comment-16644495 ] ASF GitHub Bot commented on OMID-90: GitHub user yonigottesman opened a pull request: https://github.com/apache/incubator-omid/pull/46 [OMID-90] Integrate omid low latency to phoenix changes You can merge this pull request into a Git repository by running: $ git pull https://github.com/yonigottesman/incubator-omid phoenix_ll_integration Alternatively you can review and apply these changes as the patch at: https://github.com/apache/incubator-omid/pull/46.patch To close this pull request, make a commit to your master/trunk branch with (at least) the following in the commit message: This closes #46 commit 7647733a50431f2b8f5d4495310db59932c46e06 Author: Yonatan Gottesman Date: 2018-10-10T05:49:28Z Integrate omid low latency to phoenix changes > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: Sub-task >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16590999#comment-16590999 ] Lianghong Xu commented on OMID-90: -- Hi [~yonigo], so sorry for the very late response. We haven't got a chance to work on it due to lack of resources. We'll try to get back to you in a few weeks. Thanks for your patience! Lianghong > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16577499#comment-16577499 ] Yonatan Gottesman commented on OMID-90: --- Hi [~lianghongxu], did you manage to get it working? > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16553133#comment-16553133 ] Yonatan Gottesman commented on OMID-90: --- ok, can i see you ycsb omid version? you can use ycsb to measure the latency of each stage of the transaction (begin, hbase, commit) and then share it with us. begin time should be an order of magnitude less then original omid > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16553122#comment-16553122 ] Lianghong Xu commented on OMID-90: -- Hi [~yonigo], I thought so before, but we were able to get 110K QPS with YCSB on hbase with this setup. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16553119#comment-16553119 ] Yonatan Gottesman commented on OMID-90: --- Hi [~lianghongxu], can you test the throughput of regular read/writes to *hbase* with this setup? (without omid at all). Maybe you are reaching the max throughput ycsb can get > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16553118#comment-16553118 ] Lianghong Xu commented on OMID-90: -- Hi [~yonigo], We were using ycsb 0.15.0-SNAPSHOT. The experiment setup is as follow: Hardware: 12 Omid clients: i3.2xlarge instances. 2 TSO servers: HA setup, c3.8xlarge instances 20 HBase regionservers: i3.2xlarge instances. YCSB: 1M records, 80/20 read/write ratio. uniform dist. filedcount = 4. For original Omid, we used numConcurrentCTWriters: 8 batchSizePerCTWriter: 50 The max throughput we could obtain for both original and OmidLL is around 50-65K. We tried various tuning ourselves but couldn't seem to get it higher. Thanks, Lianghong > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16553109#comment-16553109 ] Lianghong Xu commented on OMID-90: -- Hi [~yonigo], We were using ycsb 0.15.0-SNAPSHOT. The experiment setup is as follow: Hardware: 12 Omid clients: i3.2xlarge instances. 2 TSO servers: HA setup, c3.8xlarge instances 20 HBase regionservers: i3.2xlarge instances. YCSB: 1M records, 80/20 read/write ratio. uniform dist. filedcount = 4. For original Omid, we used numConcurrentCTWriters: 8 batchSizePerCTWriter: 50 The max throughput we could obtain for both original and OmidLL is around 50-65K. We tried various tuning ourselves but couldn't seem to get it higher. Thanks, Lianghong > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16547434#comment-16547434 ] Yonatan Gottesman commented on OMID-90: --- Hi [~lianghongxu], Can you please share with us the version of ycsb you are running, and the setup of your experiment? As you can see in the paper, low latency benefits are seen when running in high throughput. To what throughput do you get to with your ycsb instance? Thanks > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16543619#comment-16543619 ] Lianghong Xu commented on OMID-90: -- Hi [~ohads] [~ebortnik] I just did an initial evaluation with Omid Low Latency version with YCSB. However, the result in terms of throughput and latency seems similar to the original Omid. I guess maybe there is something I'm missing? I was using the default configs except that I set "lowLatency: true" to be true in omid-server-configuration.yml. I think that should be enough to turn on the low latency mode, right? The YCSB workload we used has a read/write ratio of 80/20 with uniform dist. Please let us know if you have any suggestions on tuning. Thanks, Lianghong > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16537305#comment-16537305 ] Lianghong Xu commented on OMID-90: -- Hi [~ohads] [~ebortnik] thanks for providing the low latency implementation and sorry for the late response. We have not yet got a chance to test it due to lack of resources since it's near end of H1. We will try to do the benchmark by the end of July and will keep you updated. Thanks! > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16536912#comment-16536912 ] Edward Bortnikov commented on OMID-90: -- [~lianghongxu], have you had a chance to play with the new protocol? Any feedback that you can share? Your experience matters as we move on to make the low-latency protocol part of Omid's next release. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16530297#comment-16530297 ] Ohad Shacham commented on OMID-90: -- Hi [~lianghongxu], Did you try the low latency implementation? Thx, Ohad > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16509534#comment-16509534 ] ASF GitHub Bot commented on OMID-90: GitHub user yonigottesman opened a pull request: https://github.com/apache/incubator-omid/pull/38 [OMID-90] Implement low latency protocol for omid. TSO does not commit transactions to commit table and lets the client write to commit table. To insure snapshot isolation, clients invalidate transactions while reading from commit table. You can merge this pull request into a Git repository by running: $ git pull https://github.com/yonigottesman/incubator-omid omid-low-latency Alternatively you can review and apply these changes as the patch at: https://github.com/apache/incubator-omid/pull/38.patch To close this pull request, make a commit to your master/trunk branch with (at least) the following in the commit message: This closes #38 commit 475f35267c892cbee42316008c80df7132f29895 Author: Yonatan Gottesman Date: 2018-06-12T11:45:48Z Low latency protocol > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Assignee: Yonatan Gottesman >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16489509#comment-16489509 ] Lianghong Xu commented on OMID-90: -- [~ebortnik] That sounds great! Please let me know when it is ready. Thanks! > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16488611#comment-16488611 ] Edward Bortnikov commented on OMID-90: -- [~lianghongxu], I work with [~ohads], nice to e-meet :). We can provide experimental code in 2-3 weeks, would that be a fit for you? Thanks. > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16488127#comment-16488127 ] Lianghong Xu commented on OMID-90: -- Hi [~ohads], just a kind reminder, do you have any near plan to open source this new version of Omid? Thanks, Lianghong > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16479432#comment-16479432 ] Lianghong Xu commented on OMID-90: -- Hi [~ohads], Very cool and congrats on the VLDB acceptance! We are evaluating Omid for our ads systems, but was hitting a performance bottleneck caused by the centralized batch commit. Do you have a plan when to open source Omid LL? Also, at this point, would it be possible for you to provide us with the Omid LL prototype so that we can evaluate its performance? Thanks! Lianghong > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Priority: Major > Attachments: OmidCloud-VLDB.pdf > > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16478884#comment-16478884 ] Ohad Shacham commented on OMID-90: -- Hi [~lianghongxu], Please find attached a paper describing the low latency version of Omid. The paper will appear in VLDB 2018. It would be very interesting to know the use case of Omid in your production system. Is it possible? Thanks, Ohad > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Priority: Major > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (OMID-90) Reducing begin/commit latency by distributing the write to the commit table
[ https://issues.apache.org/jira/browse/OMID-90?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16476306#comment-16476306 ] Lianghong Xu commented on OMID-90: -- Hi, we are exploring using Omid in our production systems at Pinterest and also observed the same issue described above. I'm wondering what is the current progress of this? If you have a design doc, would you mind sharing it? Thanks! > Reducing begin/commit latency by distributing the write to the commit table > --- > > Key: OMID-90 > URL: https://issues.apache.org/jira/browse/OMID-90 > Project: Apache Omid > Issue Type: New Feature >Reporter: Ohad Shacham >Priority: Major > > Today, Omid's commits are done by the transaction manager. In order to > efficiently write to the commit table, the transaction manager batches these > writes. This optimization, even thought reduces the write time to HBase, > significantly increases the begin and commit latency. The commit latency > increases since a commit operation returns only after its commit timestamp > was persisted in the commit table. And the begin latency increases since > begin returns a transaction id that is also used by the transaction to > identify its snapshot and therefore, begin returns only after all commits > with commit id smaller than the begin id was persisted in the commit table. > This is crucial, since a snapshot change during a transaction run may violate > snapshot isolation. > > The idea of this feature is to distribute the commit by moving the write to > the commit table from the server to the client. The transaction manager does > conflict analysis and returns a commit timestamp. While the client atomically > persists this commit in the commit table. > This significantly reduces the begin and commit latency, since batching is > not required anymore. A begin operation can immediately returns and a commit > operation returns after conflict detection. > This can introduce snapshot isolation violation since a slow client can > commit and change other transaction's snapsho. Therefore, we use an > invalidation technique which is similar to the one Omid uses today to > maintain snapshot isolation in high availability mode. -- This message was sent by Atlassian JIRA (v7.6.3#76005)