[
https://issues.apache.org/jira/browse/TEPHRA-247?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16324938#comment-16324938
]
Patrick Xiaoman Huang edited comment on TEPHRA-247 at 1/15/18 7:30 AM:
-----------------------------------------------------------------------
Hi Andreas,
As several regions may reside on region server, tx cache is not key related,
it's table related, so I think cache instance should be a regionserver observer
that is one instance for one regionserver jvm.
and we configure the HTable pool send start/update/commit tx message to region
servers spontaneously.
--region split
yes, this is the problem, can we use RegionObserver.postCompleteSplit() do
some Cache.split() means push caches to no cache region server?
– timeout
tx has attribute of timeout, we can use this to clean the tx out of cache.
– checkpoint
When checkpoint, table will update new tx to server, tx will used in
TransactionProcessor right? any problem?
Patrick
was (Author: mk926):
Hi Andres,
when we start a tx, yes we will send the request to all region,
HTable.coprocesorService use several Futures to do the invocations
concurrently, so I think this will not affect the performance, but your concern
is this will increase the probilities of failure?
--region split
yes, this is the problem, can we use RegionObserver.postCompleteSplit() do some
Cache.split() means push caches to splited region?
-- timeout
tx has attribute of timeout, we use this to clean the tx out of cache?
-- checkpoint
I will look more on this.
I will do some more thinking on these.
Patrick
Thanks for the reply
> Avoid encoding the transaction multiple times
> ---------------------------------------------
>
> Key: TEPHRA-247
> URL: https://issues.apache.org/jira/browse/TEPHRA-247
> Project: Tephra
> Issue Type: Improvement
> Components: core, manager
> Affects Versions: 0.12.0-incubating
> Reporter: Andreas Neumann
> Assignee: Andreas Neumann
> Priority: Major
> Attachments: design.jpg
>
>
> Currently, the same transaction object is encoded again and again for every
> Get performed in HBase. It would be better to cache the encoded transaction
> for the duration of the transaction and reuse it,
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
