[
https://issues.apache.org/jira/browse/HBASE-18027?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16011484#comment-16011484
]
Andrew Purtell edited comment on HBASE-18027 at 5/15/17 11:12 PM:
------------------------------------------------------------------
Ok, here is a patch that moves the discussed concerns up into the caller(s).
This is in response to feedback from [~lhofhansl]. Although inside
HBaseInterClusterReplicationEndpoint the batch of edits may be broken up into
smaller batches for parallelization, up in the callers we can't see how the
sub-batches might be partitioned. So, we have to consider the RPC request limit
with respect to the whole batch, which does simplify the change: If the
replication batch capacity will exceed the RPC request limit we should simply
use the RPC request limit as the replication batch capacity. The downside is
this is more pessimistic than if we check limits right where we are building
the sub-batches and know the size of each.
Changes:
- Where we set replicationBatchSizeCapacity in
ReplicationSourceWALReaderThread, check if replicationBatchSizeCapacity will
exceed the RPC request size limit, and if so use the request size limit instead
and warn about it with mention of relevant configuration keys.
- When building the current replication batch, check if we will exceed the
batch size capacity _before_ adding an entry to the batch. Ensure at least one
entry is added to the batch even if it means we violate quota or batch size
capacity or otherwise replication would become stuck.
- WALEntryStream needs a putBack() method to undo next() when we've decided the
entry we just iterated to will cause the batch to exceed size limits.
and minor changes to logging:
- New debug logging in HBaseInterClusterReplicationEndpoint. One new DEBUG
level line per replication batch.
- Fix trace level log that conflated total amount of data in replication
context with size of data in the worklist submitted as Replicator runnable.
- Add details to the warning logged if the number of edits replicated is
different from the number received.
All replication unit tests pass with these changes applied.
was (Author: apurtell):
Ok, here is a patch that moves the discussed concerns up into the caller(s).
This is in response to feedback from [~lhofhansl]. Although inside
HBaseInterClusterReplicationEndpoint the batch of edits may be broken up into
smaller batches for parallelization, up in the callers we can't see how the
sub-batches might be partitioned. So, we have to consider the RPC request limit
with respect to the whole batch, which does simplify the change: If the
replication batch capacity will exceed the RPC request limit we should simply
use the RPC request limit as the replication batch capacity. The downside is
this is more pessimistic than if we check limits right where we are building
the sub-batches and know the size of each.
Changes:
- Where we set replicationBatchSizeCapacity in
ReplicationSourceWALReaderThread, check if replicationBatchSizeCapacity will
exceed the RPC request size limit, and if so use the request size limit instead
and warn about it with mention of relevant configuration keys.
- When building the current replication batch, check if we will exceed the
batch size capacity _before_ adding an entry to the batch. Ensure at least one
entry is added to the batch even if it means we violate quota or batch size
capacity or otherwise replication would become stuck.
- WALEntryStream needs a putBack() method to undo next() when we've decided the
entry we just iterated to will cause the batch to exceed size limits.
and minor changes to logging:
- New debug logging in HBaseInterClusterReplicationEndpoint. One new DEBUG
level line per replication batch.
- Fix trace level log that conflated total amount of data in replication
context with size of data in the worklist submitted as Replicator runnable.
- Add details to the warning logged if the number of edits replicated is
different from the number received.
All *Replication* unit tests pass with these changes applied.
> Replication should respect RPC size limits when batching edits
> --------------------------------------------------------------
>
> Key: HBASE-18027
> URL: https://issues.apache.org/jira/browse/HBASE-18027
> Project: HBase
> Issue Type: Bug
> Components: Replication
> Affects Versions: 2.0.0, 1.4.0, 1.3.1
> Reporter: Andrew Purtell
> Assignee: Andrew Purtell
> Fix For: 2.0.0, 1.4.0, 1.3.2
>
> Attachments: HBASE-18027.patch, HBASE-18027.patch, HBASE-18027.patch
>
>
> In HBaseInterClusterReplicationEndpoint#replicate we try to replicate in
> batches. We create N lists. N is the minimum of configured replicator
> threads, number of 100-waledit batches, or number of current sinks. Every
> pending entry in the replication context is then placed in order by hash of
> encoded region name into one of these N lists. Each of the N lists is then
> sent all at once in one replication RPC. We do not test if the sum of data in
> each N list will exceed RPC size limits. This code presumes each individual
> edit is reasonably small. Not checking for aggregate size while assembling
> the lists into RPCs is an oversight and can lead to replication failure when
> that assumption is violated.
> We can fix this by generating as many replication RPC calls as we need to
> drain a list, keeping each RPC under limit, instead of assuming the whole
> list will fit in one.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
