[jira] [Commented] (HBASE-14788) Splitting a region does not support the hbase.rs.evictblocksonclose config when closing source region
[ https://issues.apache.org/jira/browse/HBASE-14788?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14999475#comment-14999475 ] Randy Fox commented on HBASE-14788: --- I tested V3 of the patch and it fixes the split issue. > Splitting a region does not support the hbase.rs.evictblocksonclose config > when closing source region > - > > Key: HBASE-14788 > URL: https://issues.apache.org/jira/browse/HBASE-14788 > Project: HBase > Issue Type: Bug > Components: regionserver >Affects Versions: 1.0.0 >Reporter: Randy Fox >Assignee: Ted Yu > Attachments: 14788-v1.txt, 14788-v2.txt, 14788-v3.txt > > > i have a table with bucket cache turned on and hbase.rs.evictblocksonclose > set to false. I split a region and watched that the closing of the source > region did not complete until the bucketcache was flushed for that region. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-14788) Splitting a region does not support the hbase.rs.evictblocksonclose config when closing source region
[ https://issues.apache.org/jira/browse/HBASE-14788?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14996885#comment-14996885 ] Randy Fox commented on HBASE-14788: --- Yes, I grepped the code and there are a few places with closeReader(true), so this may not be the last use case for this bug. > Splitting a region does not support the hbase.rs.evictblocksonclose config > when closing source region > - > > Key: HBASE-14788 > URL: https://issues.apache.org/jira/browse/HBASE-14788 > Project: HBase > Issue Type: Bug > Components: regionserver >Affects Versions: 1.0.0 >Reporter: Randy Fox > > i have a table with bucket cache turned on and hbase.rs.evictblocksonclose > set to false. I split a region and watched that the closing of the source > region did not complete until the bucketcache was flushed for that region. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Created] (HBASE-14788) Splitting a region does not support the hbase.rs.evictblocksonclose config when closing source region
Randy Fox created HBASE-14788: - Summary: Splitting a region does not support the hbase.rs.evictblocksonclose config when closing source region Key: HBASE-14788 URL: https://issues.apache.org/jira/browse/HBASE-14788 Project: HBase Issue Type: Bug Components: regionserver Affects Versions: 1.0.0 Reporter: Randy Fox i have a table with bucket cache turned on and hbase.rs.evictblocksonclose set to false. I split a region and watched that the closing of the source region did not complete until the bucketcache was flushed for that region. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-14624) BucketCache.freeBlock is too expensive
[ https://issues.apache.org/jira/browse/HBASE-14624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14971587#comment-14971587 ] Randy Fox commented on HBASE-14624: --- As a baseline (and to install the patch) I did a rolling restart of one RS and it took 7 minutes to move the regions. i waited a few hours for bucket cache to fill up and then moved all the regions off and it took 20 seconds or so and I verified the buckets freed up. > BucketCache.freeBlock is too expensive > -- > > Key: HBASE-14624 > URL: https://issues.apache.org/jira/browse/HBASE-14624 > Project: HBase > Issue Type: Bug > Components: BlockCache >Affects Versions: 1.0.0 >Reporter: Randy Fox > Fix For: 2.0.0, 1.2.0, 1.3.0, 1.0.3, 1.1.3, 0.98.16 > > Attachments: 14624-v1.txt, 14624-v2.txt > > > Moving regions is unacceptably slow when using bucket cache, as it takes too > long to free all the blocks. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-14624) BucketCache.freeBlock is too expensive
[ https://issues.apache.org/jira/browse/HBASE-14624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14970138#comment-14970138 ] Randy Fox commented on HBASE-14624: --- yes > BucketCache.freeBlock is too expensive > -- > > Key: HBASE-14624 > URL: https://issues.apache.org/jira/browse/HBASE-14624 > Project: HBase > Issue Type: Bug > Components: BlockCache >Affects Versions: 1.0.0 >Reporter: Randy Fox > Fix For: 2.0.0, 1.2.0, 1.3.0, 1.0.3, 1.1.3, 0.98.16 > > Attachments: 14624-v1.txt > > > Moving regions is unacceptably slow when using bucket cache, as it takes too > long to free all the blocks. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-14624) BucketCache.freeBlock is too expensive
[ https://issues.apache.org/jira/browse/HBASE-14624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14970070#comment-14970070 ] Randy Fox commented on HBASE-14624: --- The test passed with flying colors. Regions close super fast with this change. > BucketCache.freeBlock is too expensive > -- > > Key: HBASE-14624 > URL: https://issues.apache.org/jira/browse/HBASE-14624 > Project: HBase > Issue Type: Improvement > Components: BlockCache >Affects Versions: 1.0.0 >Reporter: Randy Fox > Attachments: 14624-v1.txt > > > Moving regions is unacceptably slow when using bucket cache, as it takes too > long to free all the blocks. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-14663) HStore::close does not honor config hbase.rs.evictblocksonclose
[ https://issues.apache.org/jira/browse/HBASE-14663?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14969425#comment-14969425 ] Randy Fox commented on HBASE-14663: --- tested in my production cluster and it seems to work fine > HStore::close does not honor config hbase.rs.evictblocksonclose > --- > > Key: HBASE-14663 > URL: https://issues.apache.org/jira/browse/HBASE-14663 > Project: HBase > Issue Type: Bug > Components: BlockCache, regionserver >Affects Versions: 1.0.0 >Reporter: Randy Fox >Assignee: Vladimir Rodionov >Priority: Minor > Fix For: 2.0.0, 1.2.0, 1.3.0, 1.0.3, 1.1.3 > > Attachments: HBASE-14663-v1.patch > > > I noticed moving regions was slow and due to the wait for the bucket cache to > clear. I tried setting hbase.rs.evictblocksonclose and it did not help. > I see the HStore::close method has evictonclose hard coded to true instead of > letting the config dictate: > // close each store file in parallel > CompletionService completionService = >new ExecutorCompletionService(storeFileCloserThreadPool); > for (final StoreFile f : result) { >completionService.submit(new Callable() { > @Override > public Void call() throws IOException { >f.closeReader(true); >return null; > } >}); > } -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-14624) BucketCache.freeBlock is too expensive
[ https://issues.apache.org/jira/browse/HBASE-14624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14969419#comment-14969419 ] Randy Fox commented on HBASE-14624: --- The region that took 3 minutes to close was 1.4G compressed. The evict blocks on close was not working (https://issues.apache.org/jira/browse/HBASE-14663). The patch for it seems to solve the problem. I can give the above change a try. > BucketCache.freeBlock is too expensive > -- > > Key: HBASE-14624 > URL: https://issues.apache.org/jira/browse/HBASE-14624 > Project: HBase > Issue Type: Improvement > Components: BlockCache >Affects Versions: 1.0.0 >Reporter: Randy Fox > Attachments: 14624-v1.txt > > > Moving regions is unacceptably slow when using bucket cache, as it takes too > long to free all the blocks. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Created] (HBASE-14663) HStore::close does not honor config hbase.rs.evictblocksonclose
Randy Fox created HBASE-14663: - Summary: HStore::close does not honor config hbase.rs.evictblocksonclose Key: HBASE-14663 URL: https://issues.apache.org/jira/browse/HBASE-14663 Project: HBase Issue Type: Bug Components: BlockCache, regionserver Affects Versions: 1.0.0 Reporter: Randy Fox Priority: Minor I noticed moving regions was slow and due to the wait for the bucket cache to clear. I tried setting hbase.rs.evictblocksonclose and it did not help. I see the HStore::close method has evictonclose hard coded to true instead of letting the config dictate: // close each store file in parallel CompletionService completionService = new ExecutorCompletionService(storeFileCloserThreadPool); for (final StoreFile f : result) { completionService.submit(new Callable() { @Override public Void call() throws IOException { f.closeReader(true); return null; } }); } -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-14624) BucketCache.freeBlock is too expensive
[ https://issues.apache.org/jira/browse/HBASE-14624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14963477#comment-14963477 ] Randy Fox commented on HBASE-14624: --- There are no logs for that. Deduced it from the code and that fact that when i added a table to cache it also started taking along time to move when it was previously quick. Vladimir Rodionov suggested i open this Jira after discussions on the maillist. > BucketCache.freeBlock is too expensive > -- > > Key: HBASE-14624 > URL: https://issues.apache.org/jira/browse/HBASE-14624 > Project: HBase > Issue Type: Improvement > Components: BlockCache >Affects Versions: 1.0.0 >Reporter: Randy Fox > > Moving regions is unacceptably slow when using bucket cache, as it takes too > long to free all the blocks. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-14624) BucketCache.freeBlock is too expensive
[ https://issues.apache.org/jira/browse/HBASE-14624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14963452#comment-14963452 ] Randy Fox commented on HBASE-14624: --- The region was about 1.4G compressed and it took about 2.5 minutes to move. The bucketcache is configed at 72G, but was not close to full. The bucket sizes: 9216,17408,33792,66560 2015-10-15 08:34:28,510 INFO org.apache.hadoop.hbase.regionserver.RSRpcServices: Close dad6c71ed395df19220ef1056a110086, moving to hb17.prod1.connexity.net,60020,125924021 2015-10-15 08:34:28,510 DEBUG org.apache.hadoop.hbase.regionserver.handler.CloseRegionHandler: Processing close of Wildfire_graph3,\x00)"y\x1B\xF0\x13-,1434402364692.dad6c71ed395df19220ef1056a110086. 2015-10-15 08:34:28,511 DEBUG org.apache.hadoop.hbase.regionserver.HRegion: Closing Wildfire_graph3,\x00)"y\x1B\xF0\x13-,1434402364692.dad6c71ed395df19220ef1056a110086.: disabling compactions & flushes 2015-10-15 08:34:28,511 INFO org.apache.hadoop.hbase.regionserver.HRegion: Running close preflush of Wildfire_graph3,\x00)"y\x1B\xF0\x13-,1434402364692.dad6c71ed395df19220ef1056a110086. 2015-10-15 08:34:28,511 INFO org.apache.hadoop.hbase.regionserver.HRegion: Started memstore flush for Wildfire_graph3,\x00)"y\x1B\xF0\x13-,1434402364692.dad6c71ed395df19220ef1056a110086., current region memstore size 44.89 MB, and 1/1 column families' memstores are being flushed. 2015-10-15 08:34:28,511 WARN org.apache.hadoop.hbase.regionserver.wal.FSHLog: Couldn't find oldest seqNum for the region we are about to flush: [dad6c71ed395df19220ef1056a110086] 2015-10-15 08:34:29,137 INFO org.apache.hadoop.hbase.regionserver.DefaultStoreFlusher: Flushed, sequenceid=70839856845, memsize=44.9 M, hasBloomFilter=false, into tmp file hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/.tmp/3600b47839a945de9733cf17581458e0 2015-10-15 08:34:29,144 DEBUG org.apache.hadoop.hbase.regionserver.HRegionFileSystem: Committing store file hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/.tmp/3600b47839a945de9733cf17581458e0 as hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/L/3600b47839a945de9733cf17581458e0 2015-10-15 08:34:29,150 INFO org.apache.hadoop.hbase.regionserver.HStore: Added hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/L/3600b47839a945de9733cf17581458e0, entries=190131, sequenceid=70839856845, filesize=3.2 M 2015-10-15 08:34:29,151 INFO org.apache.hadoop.hbase.regionserver.HRegion: Finished memstore flush of ~44.89 MB/47066832, currentsize=22.39 KB/22928 for region Wildfire_graph3,\x00)"y\x1B\xF0\x13-,1434402364692.dad6c71ed395df19220ef1056a110086. in 640ms, sequenceid=70839856845, compaction requested=false 2015-10-15 08:34:29,152 DEBUG org.apache.hadoop.hbase.regionserver.HRegion: Updates disabled for region Wildfire_graph3,\x00)"y\x1B\xF0\x13-,1434402364692.dad6c71ed395df19220ef1056a110086. 2015-10-15 08:34:29,152 INFO org.apache.hadoop.hbase.regionserver.HRegion: Started memstore flush for Wildfire_graph3,\x00)"y\x1B\xF0\x13-,1434402364692.dad6c71ed395df19220ef1056a110086., current region memstore size 22.39 KB, and 1/1 column families' memstores are being flushed. 2015-10-15 08:34:29,152 WARN org.apache.hadoop.hbase.regionserver.wal.FSHLog: Couldn't find oldest seqNum for the region we are about to flush: [dad6c71ed395df19220ef1056a110086] 2015-10-15 08:34:29,225 INFO org.apache.hadoop.hbase.regionserver.DefaultStoreFlusher: Flushed, sequenceid=70839856876, memsize=22.4 K, hasBloomFilter=false, into tmp file hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/.tmp/a8a74549c7a44174b105cbed23cfaea1 2015-10-15 08:34:29,231 DEBUG org.apache.hadoop.hbase.regionserver.HRegionFileSystem: Committing store file hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/.tmp/a8a74549c7a44174b105cbed23cfaea1 as hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/L/a8a74549c7a44174b105cbed23cfaea1 2015-10-15 08:34:29,279 INFO org.apache.hadoop.hbase.regionserver.HStore: Added hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/L/a8a74549c7a44174b105cbed23cfaea1, entries=128, sequenceid=70839856876, filesize=2.6 K 2015-10-15 08:34:29,280 INFO org.apache.hadoop.hbase.regionserver.HRegion: Finished memstore flush of ~22.39 KB/22928, currentsize=0 B/0 for region Wildfire_graph3,\x00)"y\x1B\xF0\x13-,1434402364692.dad6c71ed395df19220ef1056a110086. in 128ms, sequenceid=70839856876, compaction requested=true ... 2015-10-15 08:37:02,945 INFO org.apache.hadoop.hbase.regionserver.HStore: Closed L 2015-10-15 08:37:02,954 DEBUG org.apache.hadoop.hbase.wal.WALSplitter: Wrote region seqId=hdfs://woz/hbase/data/default/Wildfire_graph3/dad6c71ed395df19220ef1056a110086/recovered.edits/7083985
[jira] [Created] (HBASE-14624) BucketCache.freeBlock is too expensive
Randy Fox created HBASE-14624: - Summary: BucketCache.freeBlock is too expensive Key: HBASE-14624 URL: https://issues.apache.org/jira/browse/HBASE-14624 Project: HBase Issue Type: Improvement Components: BlockCache Affects Versions: 1.0.0 Reporter: Randy Fox Moving regions is unacceptably slow when using bucket cache, as it takes too long to free all the blocks. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HBASE-7272) HFileOutputFormat.configureIncrementalLoad should honor table Max File Size and Columns BlockSize
[ https://issues.apache.org/jira/browse/HBASE-7272?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13533999#comment-13533999 ] Randy Fox commented on HBASE-7272: -- configureIncrementalLoad takes an HTable as a parameter, so shouldn't it be able to make these determinations? > HFileOutputFormat.configureIncrementalLoad should honor table Max File Size > and Columns BlockSize > - > > Key: HBASE-7272 > URL: https://issues.apache.org/jira/browse/HBASE-7272 > Project: HBase > Issue Type: Improvement > Components: HFile >Affects Versions: 0.92.1 >Reporter: Randy Fox >Priority: Minor > > HFileOutputFormat.configureIncrementalLoad is used to generate HFiles > matching region assignments. The problem is that is it does not using file > and block size settings. This can create a lot of files of wrong block size, > which may take a long time to compact. I think it should honor these > settings and expedite its use. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Created] (HBASE-7272) HFileOutputFormat.configureIncrementalLoad should honor tabel Max File Size and Columns BlockSize
Randy Fox created HBASE-7272: Summary: HFileOutputFormat.configureIncrementalLoad should honor tabel Max File Size and Columns BlockSize Key: HBASE-7272 URL: https://issues.apache.org/jira/browse/HBASE-7272 Project: HBase Issue Type: Improvement Components: HFile Affects Versions: 0.92.1 Reporter: Randy Fox Priority: Minor HFileOutputFormat.configureIncrementalLoad is used to generate HFiles matching region assignments. The problem is that is it does not using file and block size settings. This can create a lot of files of wrong block size, which may take a long time to compact. I think it should honor these settings and expedite its use. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira