[
https://issues.apache.org/jira/browse/OMID-140?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16810289#comment-16810289
]
Lars Hofhansl commented on OMID-140:
------------------------------------
Thanks. Yeah there has to be some limit since the TSO needs to know the entire
write set, in order to find the intersections. (Or perhaps this is where my
idea with the Bloomfilters representing the writeset might come in handy...?)
In any case I was looking for this spot. Thanks for pointing me there. Might be
good to make this more explicit. Right now it looks like a network or a
protocol failure. Maybe throw a different exception and log loudly in the TSO
log? It was enough to confuse me :)
Should this be configurable? 10MB seems like a good default, but some people
want more, and perhaps some folks want to fail earlier... And in that case
maybe this should fail at the client even before we try to ship it to the TSO.
> ServiceUnavailableException when the write set if is too large.
> ---------------------------------------------------------------
>
> Key: OMID-140
> URL: https://issues.apache.org/jira/browse/OMID-140
> Project: Apache Omid
> Issue Type: Bug
> Reporter: Lars Hofhansl
> Priority: Major
>
> In the debugger I see this:
> {code:java}
> java.util.concurrent.ExecutionException:
> org.apache.omid.tso.client.ServiceUnavailableException: Number of retries
> exceeded. This API request failed permanently
> {code}
> {code:java}
> Thrown from here:
> HBaseTransactionManager(AbstractTransactionManager).commitRegularTransaction(AbstractTransaction<CellId>)
> line: 417
> HBaseTransactionManager(AbstractTransactionManager).commit(Transaction) line:
> 252
> OmidTransactionContext.commit() line: 124
> MutationState.commit() line: 1226
> PhoenixConnection$3.call() line: 673
> PhoenixConnection$3.call() line: 669
> CallRunner.run(T, CallWrapper...) line: 53
> PhoenixConnection.commit() line: 669
> {code}
> In my case I tried with a write set of 2^20 (a bit over 1m) rows (that came
> to 4m cells.)
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
