Re: Unique key constraint and optimistic locking (versioning)
Created SOLR-3178 covering the versioning/optimistic-locking part. In combination SOLR-3173 and SOLR-3178 should provide the features I am missing, and that I believe lots of other SOLR users will be able to benefit from. Please help shape by commenting on the Jira issues. Thanks. Per Steffensen skrev: Created SOLR-3173 on the part about making insert fail if document (with same uniqueKey) already exists. SOLR-3173 also includes to make update not insert document if not already exists - just for consistency with normal RDBMS behaviour. So basically the feature allowes you to turn on this behaviour of having database (RDBMS) semantics, and when you do you get both. Tomorrrow will create another Jira issue on the versioning/optimistic locking part. Per Steffensen skrev: Hi Does solr/lucene provide any mechanism for unique key constraint and optimistic locking (versioning)? Unique key constraint: That a client will not succeed creating a new document in solr/lucene if a document already exists having the same value in some field (e.g. an id field). Of course implemented right, so that even though two or more threads are concurrently trying to create a new document with the same value in this field, only one of them will succeed. Optimistic locking (versioning): That a client will only succeed updating a document if this updated document is based on the version of the document currently stored in solr/lucene. Implemented in the optimistic way that clients during an update have to tell which version of the document they fetched from Solr and that they therefore have used as a starting-point for their updated document. So basically having a version field on the document that clients increase by one before sending to solr for update, and some code in Solr that only makes the update succeed if the version number of the updated document is exactly one higher than the version number of the document already stored. Of course again implemented right, so that even though two or more thrads are concurrently trying to update a document, and they all have their updated document based on the current version in solr/lucene, only one of them will succeed. Or do I have to do stuff like this myself outside solr/lucene - e.g. in the client using solr. Regards, Per Steffensen
Re: Unique key constraint and optimistic locking (versioning)
Created SOLR-3173 on the part about making insert fail if document (with same uniqueKey) already exists. SOLR-3173 also includes to make update not insert document if not already exists - just for consistency with normal RDBMS behaviour. So basically the feature allowes you to turn on this behaviour of having database (RDBMS) semantics, and when you do you get both. Tomorrrow will create another Jira issue on the versioning/optimistic locking part. Per Steffensen skrev: Hi Does solr/lucene provide any mechanism for unique key constraint and optimistic locking (versioning)? Unique key constraint: That a client will not succeed creating a new document in solr/lucene if a document already exists having the same value in some field (e.g. an id field). Of course implemented right, so that even though two or more threads are concurrently trying to create a new document with the same value in this field, only one of them will succeed. Optimistic locking (versioning): That a client will only succeed updating a document if this updated document is based on the version of the document currently stored in solr/lucene. Implemented in the optimistic way that clients during an update have to tell which version of the document they fetched from Solr and that they therefore have used as a starting-point for their updated document. So basically having a version field on the document that clients increase by one before sending to solr for update, and some code in Solr that only makes the update succeed if the version number of the updated document is exactly one higher than the version number of the document already stored. Of course again implemented right, so that even though two or more thrads are concurrently trying to update a document, and they all have their updated document based on the current version in solr/lucene, only one of them will succeed. Or do I have to do stuff like this myself outside solr/lucene - e.g. in the client using solr. Regards, Per Steffensen
Re: Unique key constraint and optimistic locking (versioning)
Em skrev: Hi Per, I want an error to occur if a document with the same id already exists, when my intent is to INSERT a new document. When my intent is to UPDATE a document in solr/lucene I want the old document already in solr/lucene deleted and the new version of this document added (exactly as you explain). It will not be possible for solr/lucene to decide what to do unless I give it some information about my intent - whether it is INSERT or UPDATE semantics I want. I guess solr/lucene always give me INSERT sematics when a document with the same id does not already exist, and that it always give me UPDATE semantics when a document with the same id does exist? I cannot decide? Given that you've set a uniqueKey-field and there already exists a document with that uniqueKey, it will delete the old one and insert the new one. There is really no difference between the semantics - updates do not exist. To create a UNIQUE-constraint as you know it from a database you have to check whether a document is already in the index *or* whether it is already pending (waiting for getting flushed to the index). Fortunately Solr manages a so called pending-set with all those documents waiting for beeing flushed to disk (Solr 3.5). We are using latest greates 4.0-SNAPSHOT code, because we want to take advantage of SolrCloud stuff. Can you give a code-pointer to where I can find the pending-set stuff? Does solr use this pending-set for query responses, so that solr deliver 100% real-time search results? I think you have to write your own DirectUpdateHandler to achieve what you want on the Solr-level or to extend Lucenes IndexWriter to do it on the Lucene-Level. While doing so, keep track of what is going on in the trunk and how Near-Real-Time-Search will change the current way of handling updates. Will do. We already use auto soft commits. There is not built-in way to make solr/lucene give me an error if I try to insert a new document with an id equal to a document already in the index/core/shard. The existing document will always be updated (implemented as old deleted and new added). Correct? Exactly. If you really want to get your hands on that topic I suggest you to learn more about Lucene's IndexWriter: http://lucene.apache.org/core/old_versioned_docs/versions/3_5_0/api/all/index.html?org/apache/lucene/index/IndexWriter.html Kind Regards, Em
Re: Unique key constraint and optimistic locking (versioning)
Given that you've set a uniqueKey-field and there already exists a document with that uniqueKey, it will delete the old one and insert the new one. There is really no difference between the semantics - updates do not exist. To create a UNIQUE-constraint as you know it from a database you have to check whether a document is already in the index *or* whether it is already pending (waiting for getting flushed to the index). Fortunately Solr manages a so called pending-set with all those documents waiting for beeing flushed to disk (Solr 3.5). We are using latest greates 4.0-SNAPSHOT code, because we want to take advantage of SolrCloud stuff. Can you give a code-pointer to where I can find the pending-set stuff? I am not sure if this is what you're asking but you should be able to get the latest data from Solr by using realtime get http://wiki.apache.org/solr/RealTimeGet -- Sami Siren
Re: Unique key constraint and optimistic locking (versioning)
Sami Siren skrev: Given that you've set a uniqueKey-field and there already exists a document with that uniqueKey, it will delete the old one and insert the new one. There is really no difference between the semantics - updates do not exist. To create a UNIQUE-constraint as you know it from a database you have to check whether a document is already in the index *or* whether it is already pending (waiting for getting flushed to the index). Fortunately Solr manages a so called pending-set with all those documents waiting for beeing flushed to disk (Solr 3.5). We are using latest greates 4.0-SNAPSHOT code, because we want to take advantage of SolrCloud stuff. Can you give a code-pointer to where I can find the pending-set stuff? I am not sure if this is what you're asking but you should be able to get the latest data from Solr by using realtime get http://wiki.apache.org/solr/RealTimeGet Thanks a lot! I might be very usefull, if this provide 100% real time get - that is, if it gets the latest version of the document, also when neither a soft-commit nor a hard-commit has been performed since the lastest version of the document was indexed. Does it do that, or does it need a soft commit (then I believe it is only a near real time get operation)? -- Sami Siren
Re: Unique key constraint and optimistic locking (versioning)
On Fri, Feb 24, 2012 at 12:06 PM, Per Steffensen st...@designware.dk wrote: Sami Siren skrev: Given that you've set a uniqueKey-field and there already exists a document with that uniqueKey, it will delete the old one and insert the new one. There is really no difference between the semantics - updates do not exist. To create a UNIQUE-constraint as you know it from a database you have to check whether a document is already in the index *or* whether it is already pending (waiting for getting flushed to the index). Fortunately Solr manages a so called pending-set with all those documents waiting for beeing flushed to disk (Solr 3.5). We are using latest greates 4.0-SNAPSHOT code, because we want to take advantage of SolrCloud stuff. Can you give a code-pointer to where I can find the pending-set stuff? I am not sure if this is what you're asking but you should be able to get the latest data from Solr by using realtime get http://wiki.apache.org/solr/RealTimeGet Thanks a lot! I might be very usefull, if this provide 100% real time get - that is, if it gets the latest version of the document, also when neither a soft-commit nor a hard-commit has been performed since the lastest version of the document was indexed. Does it do that, or does it need a soft commit (then I believe it is only a near real time get operation)? I believe it does not require any kind of commit to happen so it should really be a real time get as the name suggests. -- Sami Siren
Re: Unique key constraint and optimistic locking (versioning)
Hi Per, Can you give a code-pointer to where I can find the pending-set stuff? Does solr use this pending-set for query responses, so that solr deliver 100% real-time search results? As of Solr 3.5 it can be found within the DirectUpdateHandler and DirectUpdateHandler2-classes. I am currently unaware of how things change in 4.0. Kind regards, Em
Re: Unique key constraint and optimistic locking (versioning)
This is a really cool feature! Thanks for pointing us in that direction! As the Quick Start says, a document does not need a commit nor a soft-commit or anything else to be available via RealTimeGet. However, regarding a versioning-system, one always has to keep in mind that an uncommited document is not guaranteed to be persisted in the index. So if you give a Duplicate-Key-Error, because there is a pending document with that key and afterwards the server goes down for any reason, you might end up without that document inside of Solr. You need a log for failover. Kind regards, Em Am 24.02.2012 11:06, schrieb Per Steffensen: Sami Siren skrev: Given that you've set a uniqueKey-field and there already exists a document with that uniqueKey, it will delete the old one and insert the new one. There is really no difference between the semantics - updates do not exist. To create a UNIQUE-constraint as you know it from a database you have to check whether a document is already in the index *or* whether it is already pending (waiting for getting flushed to the index). Fortunately Solr manages a so called pending-set with all those documents waiting for beeing flushed to disk (Solr 3.5). We are using latest greates 4.0-SNAPSHOT code, because we want to take advantage of SolrCloud stuff. Can you give a code-pointer to where I can find the pending-set stuff? I am not sure if this is what you're asking but you should be able to get the latest data from Solr by using realtime get http://wiki.apache.org/solr/RealTimeGet Thanks a lot! I might be very usefull, if this provide 100% real time get - that is, if it gets the latest version of the document, also when neither a soft-commit nor a hard-commit has been performed since the lastest version of the document was indexed. Does it do that, or does it need a soft commit (then I believe it is only a near real time get operation)? -- Sami Siren
Re: Unique key constraint and optimistic locking (versioning)
On Fri, Feb 24, 2012 at 6:55 AM, Em mailformailingli...@yahoo.de wrote: However, regarding a versioning-system, one always has to keep in mind that an uncommited document is not guaranteed to be persisted in the index. We now have durability via an update log. With a recent nightly trunk build, you can send a document to solr w/o committing, then kill -9 the JVM, then restart it and the log will be used to recover that document (and you should be able to see it in the index) -Yonik lucidimagination.com
Re: Unique key constraint and optimistic locking (versioning)
Em skrev: This is a really cool feature! Thanks for pointing us in that direction! A feature where you can flag your index operation to provide create sematics would be cool. When setting the create-semantics flag, an index operation will fail if a document with simular id (or whatever you use for uniqueKey) already exist. When the flag is not set index semantics will be just as it is today. ElasticSearch has this, except that they call it OpType which has the possible values create and index (index is default). Most other alternatives to Solr(Cloud) provide this as well. We need it in my current project. We might make it outside Solr/Lucene but I hope to be able to convince my ProductOwner to make it as a Solr-feature contributing it back - especiallly if the Solr community agrees that it would be a nice and commonly usable feature. Believe it is a commonly usable feature - especially when using Solr as a NoSQL data store and not just a search index (as http://wiki.apache.org/solr/RealTimeGet says) As the Quick Start says, a document does not need a commit nor a soft-commit or anything else to be available via RealTimeGet. However, regarding a versioning-system, one always has to keep in mind that an uncommited document is not guaranteed to be persisted in the index. So if you give a Duplicate-Key-Error, because there is a pending document with that key and afterwards the server goes down for any reason, you might end up without that document inside of Solr. You need a log for failover. Yes I know. Or you might just not consider a datarecord inserted into Solr before it has been indexed AND a hard-commit has happened. You can have many threads working on indexing datarecords into Solr but not deleting/acknowledging the source for those datarecords before next hard-commit has happend after index. But I believe it is another issue - one we also have plans about dealing with. Thanks everybody! Kind regards, Em Am 24.02.2012 11:06, schrieb Per Steffensen: Sami Siren skrev: Given that you've set a uniqueKey-field and there already exists a document with that uniqueKey, it will delete the old one and insert the new one. There is really no difference between the semantics - updates do not exist. To create a UNIQUE-constraint as you know it from a database you have to check whether a document is already in the index *or* whether it is already pending (waiting for getting flushed to the index). Fortunately Solr manages a so called pending-set with all those documents waiting for beeing flushed to disk (Solr 3.5). We are using latest greates 4.0-SNAPSHOT code, because we want to take advantage of SolrCloud stuff. Can you give a code-pointer to where I can find the pending-set stuff? I am not sure if this is what you're asking but you should be able to get the latest data from Solr by using realtime get http://wiki.apache.org/solr/RealTimeGet Thanks a lot! I might be very usefull, if this provide 100% real time get - that is, if it gets the latest version of the document, also when neither a soft-commit nor a hard-commit has been performed since the lastest version of the document was indexed. Does it do that, or does it need a soft commit (then I believe it is only a near real time get operation)? -- Sami Siren
Re: Unique key constraint and optimistic locking (versioning)
Yonik Seeley skrev: On Fri, Feb 24, 2012 at 6:55 AM, Em mailformailingli...@yahoo.de wrote: However, regarding a versioning-system, one always has to keep in mind that an uncommited document is not guaranteed to be persisted in the index. We now have durability via an update log. With a recent nightly trunk build, you can send a document to solr w/o committing, then kill -9 the JVM, then restart it and the log will be used to recover that document (and you should be able to see it in the index) Cool. We have a test doing exactly that - indexing 2000 documents into Solr, kill-9'ing Solr in the middle of the process, starting Solr again and checking that 2000 documents will eventually be searchable. It lights red as it is right now, but we are using a 4.0-SNAPSHOT from late december. We will try to update to newest code and see if it lights green :-) Do you have to do something to enable this log-and-recover feature, or does it just run out-of-the-box? Any documentation? -Yonik lucidimagination.com
Re: Unique key constraint and optimistic locking (versioning)
On Fri, Feb 24, 2012 at 9:04 AM, Per Steffensen st...@designware.dk wrote: Cool. We have a test doing exactly that - indexing 2000 documents into Solr, kill-9'ing Solr in the middle of the process, starting Solr again and checking that 2000 documents will eventually be searchable. It lights red as it is right now, but we are using a 4.0-SNAPSHOT from late december. We will try to update to newest code and see if it lights green :-) Do you have to do something to enable this log-and-recover feature, or does it just run out-of-the-box? Any documentation? Same as realtime-get, you need the update log configured. It's currently tested in TestRecovery. -Yonik lucidimagination.com
Re: Unique key constraint and optimistic locking (versioning)
Yonik Seeley skrev: On Fri, Feb 24, 2012 at 9:04 AM, Per Steffensen st...@designware.dk wrote: Cool. We have a test doing exactly that - indexing 2000 documents into Solr, kill-9'ing Solr in the middle of the process, starting Solr again and checking that 2000 documents will eventually be searchable. It lights red as it is right now, but we are using a 4.0-SNAPSHOT from late december. We will try to update to newest code and see if it lights green :-) Do you have to do something to enable this log-and-recover feature, or does it just run out-of-the-box? Any documentation? Same as realtime-get, you need the update log configured. It's currently tested in TestRecovery. Thanks! Any performance measurements comparing with and without update log? -Yonik lucidimagination.com
Re: Unique key constraint and optimistic locking (versioning)
Per Steffensen skrev: Em skrev: This is a really cool feature! Thanks for pointing us in that direction! A feature where you can flag your index operation to provide create sematics would be cool. When setting the create-semantics flag, an index operation will fail if a document with simular id (or whatever you use for uniqueKey) already exist. When the flag is not set index semantics will be just as it is today. ElasticSearch has this, except that they call it OpType which has the possible values create and index (index is default). Most other alternatives to Solr(Cloud) provide this as well. We need it in my current project. We might make it outside Solr/Lucene but I hope to be able to convince my ProductOwner to make it as a Solr-feature contributing it back - especiallly if the Solr community agrees that it would be a nice and commonly usable feature. Believe it is a commonly usable feature - especially when using Solr as a NoSQL data store and not just a search index (as http://wiki.apache.org/solr/RealTimeGet says) But of course if you guys who know Solr/Lucene source will make if for us, it will be greatly apprechiated :-)
Re: Unique key constraint and optimistic locking (versioning)
On Fri, Feb 24, 2012 at 8:59 AM, Per Steffensen st...@designware.dk wrote: We might make it outside Solr/Lucene but I hope to be able to convince my ProductOwner to make it as a Solr-feature contributing it back - especiallly if the Solr community agrees that it would be a nice and commonly usable feature. Our current distributed indexing (solr cloud) design explicitly took optimistic concurrency into account, and it's been on my todo list. It should actually be pretty easy - we have all the plumbing in place now that we already use for distributed indexing. For example, versioning is already used to handle reorders of updates when a leader forwards updates to replicas (and an older update to a document is simply dropped). If you'd like to help out, the starting point where versioning is handled is in DistributedUpdateProcessor (which is now a default processor and also works in non-distrib mode). -Yonik lucidimagination.com
Re: Unique key constraint and optimistic locking (versioning)
Hi Per, if you are evaluating with your ProductOwner whether he/she wants to contribute back: Try to not see it only as a gift to the community for a highly usefull product, but also see it as a protection of your investment. What you are going to customize will be deeply integrated in Solr - in code where changes might occur more often than in others, since they are critical to Solr's overall performance. When you are contributing back and committers rate your contribution as valuable and safe enough to commit it to the trunk, almost every enhancement of those code-regions is likely to be compatible with your changes. This highly reduces maintenance-costs, while it leaves more ressources for new innovations and features that make your product great. Kind regards, Em We might make it outside Solr/Lucene but I hope to be able to convince my ProductOwner to make it as a Solr-feature contributing it back - especiallly if the Solr community agrees that it would be a nice and commonly usable feature. Believe it is a commonly usable feature - especially when using Solr as a NoSQL data store and not just a search index (as http://wiki.apache.org/solr/RealTimeGet says)
Re: Unique key constraint and optimistic locking (versioning)
Yonik, thanks for sharing deeper details about how SolrCloud is going to work. Do you plan to release any wiki-updates about the small details, so that other developers are able to get in touch with what you've already done there? I think small guides and the mentioning of class-names and their relationships and/or lifecycles are good startpoints that do not need much time to write them down (take How to write distributed SearchComponents as an example). Kind regards, Em Am 24.02.2012 15:18, schrieb Yonik Seeley: On Fri, Feb 24, 2012 at 8:59 AM, Per Steffensen st...@designware.dk wrote: We might make it outside Solr/Lucene but I hope to be able to convince my ProductOwner to make it as a Solr-feature contributing it back - especiallly if the Solr community agrees that it would be a nice and commonly usable feature. Our current distributed indexing (solr cloud) design explicitly took optimistic concurrency into account, and it's been on my todo list. It should actually be pretty easy - we have all the plumbing in place now that we already use for distributed indexing. For example, versioning is already used to handle reorders of updates when a leader forwards updates to replicas (and an older update to a document is simply dropped). If you'd like to help out, the starting point where versioning is handled is in DistributedUpdateProcessor (which is now a default processor and also works in non-distrib mode). -Yonik lucidimagination.com
Re: Unique key constraint and optimistic locking (versioning)
On Feb 24, 2012, at 6:55 AM, Em wrote: You need a log for failover. There is a transaction log. - Mark Miller lucidimagination.com
Re: Unique key constraint and optimistic locking (versioning)
Em skrev: Hi Per, Solr provides the so called UniqueKey-field. Refer to the Wiki to learn more: http://wiki.apache.org/solr/UniqueKey Belive the uniqueKey does not enforce a unique key constraint, so that you are not allowed to create a document with an id's when an document with the same id already exists. So it is not the whole solution. Optimistic locking (versioning) ... is not provided by Solr out of the box. If you add a new document with the same UniqueKey it replaces the old one. You have to do the versioning on your own (and keep in mind concurrent updates). Kind regards, Em Am 21.02.2012 13:50, schrieb Per Steffensen: Hi Does solr/lucene provide any mechanism for unique key constraint and optimistic locking (versioning)? Unique key constraint: That a client will not succeed creating a new document in solr/lucene if a document already exists having the same value in some field (e.g. an id field). Of course implemented right, so that even though two or more threads are concurrently trying to create a new document with the same value in this field, only one of them will succeed. Optimistic locking (versioning): That a client will only succeed updating a document if this updated document is based on the version of the document currently stored in solr/lucene. Implemented in the optimistic way that clients during an update have to tell which version of the document they fetched from Solr and that they therefore have used as a starting-point for their updated document. So basically having a version field on the document that clients increase by one before sending to solr for update, and some code in Solr that only makes the update succeed if the version number of the updated document is exactly one higher than the version number of the document already stored. Of course again implemented right, so that even though two or more thrads are concurrently trying to update a document, and they all have their updated document based on the current version in solr/lucene, only one of them will succeed. Or do I have to do stuff like this myself outside solr/lucene - e.g. in the client using solr. Regards, Per Steffensen
Re: Unique key constraint and optimistic locking (versioning)
Hi Per, well, Solr has no Update-Method like a RDBMS. It is a re-insert of the whole document. Therefore a document with an existing UniqueKey marks the old document as deleted and inserts the new one. However this is not the whole story, since this constraint only works per index/SolrCore/Shard (depending on your use-case). Does this help you? Kind regards, Em Am 23.02.2012 15:34, schrieb Per Steffensen: Em skrev: Hi Per, Solr provides the so called UniqueKey-field. Refer to the Wiki to learn more: http://wiki.apache.org/solr/UniqueKey Belive the uniqueKey does not enforce a unique key constraint, so that you are not allowed to create a document with an id's when an document with the same id already exists. So it is not the whole solution. Optimistic locking (versioning) ... is not provided by Solr out of the box. If you add a new document with the same UniqueKey it replaces the old one. You have to do the versioning on your own (and keep in mind concurrent updates). Kind regards, Em Am 21.02.2012 13:50, schrieb Per Steffensen: Hi Does solr/lucene provide any mechanism for unique key constraint and optimistic locking (versioning)? Unique key constraint: That a client will not succeed creating a new document in solr/lucene if a document already exists having the same value in some field (e.g. an id field). Of course implemented right, so that even though two or more threads are concurrently trying to create a new document with the same value in this field, only one of them will succeed. Optimistic locking (versioning): That a client will only succeed updating a document if this updated document is based on the version of the document currently stored in solr/lucene. Implemented in the optimistic way that clients during an update have to tell which version of the document they fetched from Solr and that they therefore have used as a starting-point for their updated document. So basically having a version field on the document that clients increase by one before sending to solr for update, and some code in Solr that only makes the update succeed if the version number of the updated document is exactly one higher than the version number of the document already stored. Of course again implemented right, so that even though two or more thrads are concurrently trying to update a document, and they all have their updated document based on the current version in solr/lucene, only one of them will succeed. Or do I have to do stuff like this myself outside solr/lucene - e.g. in the client using solr. Regards, Per Steffensen
Re: Unique key constraint and optimistic locking (versioning)
Em skrev: Hi Per, well, Solr has no Update-Method like a RDBMS. It is a re-insert of the whole document. Therefore a document with an existing UniqueKey marks the old document as deleted and inserts the new one. Yes I understand. But it is not always what I want to acheive. I want an error to occur if a document with the same id already exists, when my intent is to INSERT a new document. When my intent is to UPDATE a document in solr/lucene I want the old document already in solr/lucene deleted and the new version of this document added (exactly as you explain). It will not be possible for solr/lucene to decide what to do unless I give it some information about my intent - whether it is INSERT or UPDATE semantics I want. I guess solr/lucene always give me INSERT sematics when a document with the same id does not already exist, and that it always give me UPDATE semantics when a document with the same id does exist? I cannot decide? However this is not the whole story, since this constraint only works per index/SolrCore/Shard (depending on your use-case). Yes I know. But with the right routing strategy based on id's I will be able to acheive what I want if the feature was just there per index/core/shard. Does this help you? Yes it helps me getting sure, that what I am looking for is not there. There is not built-in way to make solr/lucene give me an error if I try to insert a new document with an id equal to a document already in the index/core/shard. The existing document will always be updated (implemented as old deleted and new added). Correct? Kind regards, Em Regards, Per Steffensen
Re: Unique key constraint and optimistic locking (versioning)
Per: Yep, you've got it. You could write a custom update handler that queried (via TermDocs or something) for the ID when your intent was to INSERT, but it'll have to be custom work. I suppose you could query with a divide-and-conquer approach, that is query for id:(1 2 58 90... all your insert IDs) and go/no-go based on whether your return had any hits, but that supposed you have some idea whether pre-existing documents are likely. But Solr doesn't have anything like you're looking for. Best Erick On Thu, Feb 23, 2012 at 10:32 AM, Per Steffensen st...@designware.dk wrote: Em skrev: Hi Per, well, Solr has no Update-Method like a RDBMS. It is a re-insert of the whole document. Therefore a document with an existing UniqueKey marks the old document as deleted and inserts the new one. Yes I understand. But it is not always what I want to acheive. I want an error to occur if a document with the same id already exists, when my intent is to INSERT a new document. When my intent is to UPDATE a document in solr/lucene I want the old document already in solr/lucene deleted and the new version of this document added (exactly as you explain). It will not be possible for solr/lucene to decide what to do unless I give it some information about my intent - whether it is INSERT or UPDATE semantics I want. I guess solr/lucene always give me INSERT sematics when a document with the same id does not already exist, and that it always give me UPDATE semantics when a document with the same id does exist? I cannot decide? However this is not the whole story, since this constraint only works per index/SolrCore/Shard (depending on your use-case). Yes I know. But with the right routing strategy based on id's I will be able to acheive what I want if the feature was just there per index/core/shard. Does this help you? Yes it helps me getting sure, that what I am looking for is not there. There is not built-in way to make solr/lucene give me an error if I try to insert a new document with an id equal to a document already in the index/core/shard. The existing document will always be updated (implemented as old deleted and new added). Correct? Kind regards, Em Regards, Per Steffensen
Re: Unique key constraint and optimistic locking (versioning)
Hi Per, I want an error to occur if a document with the same id already exists, when my intent is to INSERT a new document. When my intent is to UPDATE a document in solr/lucene I want the old document already in solr/lucene deleted and the new version of this document added (exactly as you explain). It will not be possible for solr/lucene to decide what to do unless I give it some information about my intent - whether it is INSERT or UPDATE semantics I want. I guess solr/lucene always give me INSERT sematics when a document with the same id does not already exist, and that it always give me UPDATE semantics when a document with the same id does exist? I cannot decide? Given that you've set a uniqueKey-field and there already exists a document with that uniqueKey, it will delete the old one and insert the new one. There is really no difference between the semantics - updates do not exist. To create a UNIQUE-constraint as you know it from a database you have to check whether a document is already in the index *or* whether it is already pending (waiting for getting flushed to the index). Fortunately Solr manages a so called pending-set with all those documents waiting for beeing flushed to disk (Solr 3.5). I think you have to write your own DirectUpdateHandler to achieve what you want on the Solr-level or to extend Lucenes IndexWriter to do it on the Lucene-Level. While doing so, keep track of what is going on in the trunk and how Near-Real-Time-Search will change the current way of handling updates. There is not built-in way to make solr/lucene give me an error if I try to insert a new document with an id equal to a document already in the index/core/shard. The existing document will always be updated (implemented as old deleted and new added). Correct? Exactly. If you really want to get your hands on that topic I suggest you to learn more about Lucene's IndexWriter: http://lucene.apache.org/core/old_versioned_docs/versions/3_5_0/api/all/index.html?org/apache/lucene/index/IndexWriter.html Kind Regards, Em
Re: Unique key constraint and optimistic locking (versioning)
Thanks a lot. We will use the UniqueKey feature and build versioning ourselves. Do you think it would be a good idea if we built a versioning feature into Solr/Lucene instead of doing it outside, so that others can benefit from the feature as well? Guess contributions will be made according to http://wiki.apache.org/solr/HowToContribute. It is possible for outsiders (like us) to get a SVN branch at svn.apache.org to prepare contributions, or do we have to use our own SVN? Are there any plans migrating lucene/solr codebase to Git, which will make it easier getting a separate area to work on the code (making a Git fork), and suggest the contribution back to core lucene/solr (doing a Git pull request)? Thanks! Per Steffensen Em skrev: Hi Per, Solr provides the so called UniqueKey-field. Refer to the Wiki to learn more: http://wiki.apache.org/solr/UniqueKey Optimistic locking (versioning) ... is not provided by Solr out of the box. If you add a new document with the same UniqueKey it replaces the old one. You have to do the versioning on your own (and keep in mind concurrent updates). Kind regards, Em Am 21.02.2012 13:50, schrieb Per Steffensen: Hi Does solr/lucene provide any mechanism for unique key constraint and optimistic locking (versioning)? Unique key constraint: That a client will not succeed creating a new document in solr/lucene if a document already exists having the same value in some field (e.g. an id field). Of course implemented right, so that even though two or more threads are concurrently trying to create a new document with the same value in this field, only one of them will succeed. Optimistic locking (versioning): That a client will only succeed updating a document if this updated document is based on the version of the document currently stored in solr/lucene. Implemented in the optimistic way that clients during an update have to tell which version of the document they fetched from Solr and that they therefore have used as a starting-point for their updated document. So basically having a version field on the document that clients increase by one before sending to solr for update, and some code in Solr that only makes the update succeed if the version number of the updated document is exactly one higher than the version number of the document already stored. Of course again implemented right, so that even though two or more thrads are concurrently trying to update a document, and they all have their updated document based on the current version in solr/lucene, only one of them will succeed. Or do I have to do stuff like this myself outside solr/lucene - e.g. in the client using solr. Regards, Per Steffensen
Re: Unique key constraint and optimistic locking (versioning)
Per Steffensen skrev: Thanks a lot. We will use the UniqueKey feature and build versioning ourselves. Do you think it would be a good idea if we built a versioning feature into Solr/Lucene instead of doing it outside, so that others can benefit from the feature as well? Guess contributions will be made according to http://wiki.apache.org/solr/HowToContribute. It is possible for outsiders (like us) to get a SVN branch at svn.apache.org to prepare contributions, or do we have to use our own SVN? Are there any plans migrating lucene/solr codebase to Git, which will make it easier getting a separate area to work on the code (making a Git fork), and suggest the contribution back to core lucene/solr (doing a Git pull request)? Sorry - didnt see the Eclipse (using Git) chapter on http://wiki.apache.org/solr/HowToContribute. We might contribute in that area. Thanks! Per Steffensen
Re: Unique key constraint and optimistic locking (versioning)
Hi Per, Solr provides the so called UniqueKey-field. Refer to the Wiki to learn more: http://wiki.apache.org/solr/UniqueKey Optimistic locking (versioning) ... is not provided by Solr out of the box. If you add a new document with the same UniqueKey it replaces the old one. You have to do the versioning on your own (and keep in mind concurrent updates). Kind regards, Em Am 21.02.2012 13:50, schrieb Per Steffensen: Hi Does solr/lucene provide any mechanism for unique key constraint and optimistic locking (versioning)? Unique key constraint: That a client will not succeed creating a new document in solr/lucene if a document already exists having the same value in some field (e.g. an id field). Of course implemented right, so that even though two or more threads are concurrently trying to create a new document with the same value in this field, only one of them will succeed. Optimistic locking (versioning): That a client will only succeed updating a document if this updated document is based on the version of the document currently stored in solr/lucene. Implemented in the optimistic way that clients during an update have to tell which version of the document they fetched from Solr and that they therefore have used as a starting-point for their updated document. So basically having a version field on the document that clients increase by one before sending to solr for update, and some code in Solr that only makes the update succeed if the version number of the updated document is exactly one higher than the version number of the document already stored. Of course again implemented right, so that even though two or more thrads are concurrently trying to update a document, and they all have their updated document based on the current version in solr/lucene, only one of them will succeed. Or do I have to do stuff like this myself outside solr/lucene - e.g. in the client using solr. Regards, Per Steffensen