[
https://issues.apache.org/jira/browse/SOLR-16799?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17728068#comment-17728068
]
Alex Deparvu edited comment on SOLR-16799 at 6/8/23 8:54 PM:
-------------------------------------------------------------
I did some minor manual testing on this upgrade.I even ran it on jdk20 with the
new lucene jdk memory segment api! I did not see any obvious errors so far. my
goal was to run the exiting `CloudIndexing` benchmark but gradle will not run
on jdk20 yet, so this is not possible yet.
[edit] I was able to run benchmarks on jdk19, updated the numbers I posted
initially, they were off by a lot for some reason (my mistake running
benchmarks probably)
I did have to manually remove `-XX:-UseBiasedLocking` as jdk19 will not accept
it anymore
Params common across both runs:
{noformat}
running JMH with args: CloudIndexing
# VM version: JDK 19.0.2, OpenJDK 64-Bit Server VM, 19.0.2+7-44
# Warmup: 3 iterations, 10 s each
# Measurement: 4 iterations, 20 s each
# Timeout: 300 s per iteration
# Benchmark mode: Throughput, ops/time
# Benchmark: org.apache.solr.bench.index.CloudIndexing.indexLargeDoc
{noformat}
_Using MemorySegmentIndexInput with Java 19_
{noformat}
./jmh.sh CloudIndexing -t 1 -p numReplicas=3 -p useStringUtf8Over=0 -p
directBuffer=true
{noformat}
* 1 thread
{noformat}
Benchmark (directBuffer) (nodeCount) (numReplicas)
(numShards) (scale) (useStringUtf8Over) Mode Cnt Score Error Units
CloudIndexing.indexLargeDoc true 4 3
5 1 0 thrpt 4 174.928 ± 127.040 ops/s
CloudIndexing.indexSmallDoc true 4 3
5 1 0 thrpt 4 2396.760 ± 732.884 ops/s
{noformat}
* 4 threads
{noformat}
Benchmark (directBuffer) (nodeCount) (numReplicas)
(numShards) (scale) (useStringUtf8Over) Mode Cnt Score Error
Units
CloudIndexing.indexLargeDoc true 4 3
5 1 0 thrpt 4 366.884 ± 209.291 ops/s
CloudIndexing.indexSmallDoc true 4 3
5 1 0 thrpt 4 6423.951 ± 3451.908 ops/s
{noformat}
_Disabled MemorySegmentIndexInput_
{noformat}
./jmh.sh CloudIndexing -t 1 -p numReplicas=3 -p useStringUtf8Over=0 -p
directBuffer=true -jvmArgsAppend
-Dorg.apache.lucene.store.MMapDirectory.enableMemorySegments=false
{noformat}
* 1 thread
{noformat}
Benchmark (directBuffer) (nodeCount) (numReplicas)
(numShards) (scale) (useStringUtf8Over) Mode Cnt Score Error Units
CloudIndexing.indexLargeDoc true 4 3
5 1 0 thrpt 4 174.600 ± 126.784 ops/s
CloudIndexing.indexSmallDoc true 4 3
5 1 0 thrpt 4 2366.407 ± 681.313 ops/s
{noformat}
* 4 threads
{noformat}
Benchmark (directBuffer) (nodeCount) (numReplicas)
(numShards) (scale) (useStringUtf8Over) Mode Cnt Score Error
Units
CloudIndexing.indexLargeDoc true 4 3
5 1 0 thrpt 4 383.462 ± 213.578 ops/s
CloudIndexing.indexSmallDoc true 4 3
5 1 0 thrpt 4 6495.806 ± 2936.821 ops/s
{noformat}
was (Author: alex.parvulescu):
I did some minor manual testing on this upgrade.I even ran it on jdk20 with the
new lucene jdk memory segment api! I did not see any obvious errors so far. my
goal was to run the exiting `CloudIndexing` benchmark but gradle will not run
on jdk20 yet, so this is not possible yet.
[edit] I was able to run benchmarks on jdk19, updated the numbers I posted
initially, they were off by a lot for some reason (my mistake running
benchmarks probably)
I did have to manually remove `-XX:-UseBiasedLocking` as jdk19 will not accept
it anymore
Params common across both runs:
{noformat}
running JMH with args: CloudIndexing
# VM version: JDK 19.0.2, OpenJDK 64-Bit Server VM, 19.0.2+7-44
# Warmup: 3 iterations, 10 s each
# Measurement: 4 iterations, 20 s each
# Timeout: 300 s per iteration
# Benchmark mode: Throughput, ops/time
# Benchmark: org.apache.solr.bench.index.CloudIndexing.indexLargeDoc
{noformat}
_Using MemorySegmentIndexInput with Java 19_
{noformat}
./jmh.sh CloudIndexing -t 1 -p numReplicas=3 -p useStringUtf8Over=0 -p
directBuffer=true
{noformat}
* 1 thread
{noformat}
Benchmark (directBuffer) (nodeCount) (numReplicas)
(numShards) (scale) (useStringUtf8Over) Mode Cnt Score Error Units
CloudIndexing.indexLargeDoc true 4 3
5 1 0 thrpt 4 174.928 ± 127.040 ops/s
CloudIndexing.indexSmallDoc true 4 3
5 1 0 thrpt 4 2396.760 ± 732.884 ops/s
{noformat}
* 4 threads
{noformat}
Benchmark (directBuffer) (nodeCount) (numReplicas)
(numShards) (scale) (useStringUtf8Over) Mode Cnt Score Error
Units
CloudIndexing.indexLargeDoc true 4 3
5 1 0 thrpt 4 366.884 ± 209.291 ops/s
CloudIndexing.indexSmallDoc true 4 3
5 1 0 thrpt 4 6423.951 ± 3451.908 ops/s
{noformat}
_Disabled MemorySegmentIndexInput _
{noformat}
./jmh.sh CloudIndexing -t 1 -p numReplicas=3 -p useStringUtf8Over=0 -p
directBuffer=true -jvmArgsAppend
-Dorg.apache.lucene.store.MMapDirectory.enableMemorySegments=false
{noformat}
* 1 thread
{noformat}
Benchmark (directBuffer) (nodeCount) (numReplicas)
(numShards) (scale) (useStringUtf8Over) Mode Cnt Score Error Units
CloudIndexing.indexLargeDoc true 4 3
5 1 0 thrpt 4 174.600 ± 126.784 ops/s
CloudIndexing.indexSmallDoc true 4 3
5 1 0 thrpt 4 2366.407 ± 681.313 ops/s
{noformat}
* 4 threads
{noformat}
Benchmark (directBuffer) (nodeCount) (numReplicas)
(numShards) (scale) (useStringUtf8Over) Mode Cnt Score Error
Units
CloudIndexing.indexLargeDoc true 4 3
5 1 0 thrpt 4 383.462 ± 213.578 ops/s
CloudIndexing.indexSmallDoc true 4 3
5 1 0 thrpt 4 6495.806 ± 2936.821 ops/s
{noformat}
> upgrade Solr to use Lucene 9.6.0
> --------------------------------
>
> Key: SOLR-16799
> URL: https://issues.apache.org/jira/browse/SOLR-16799
> Project: Solr
> Issue Type: Task
> Reporter: Christine Poerschke
> Priority: Major
> Labels: newdev
> Time Spent: 40m
> Remaining Estimate: 0h
>
> Apache Lucene 9.6.0 was recently released:
> [https://lists.apache.org/thread/d4zw7o3t0g4h329wk7vc7s5730bsyk5t]
> This ticket is to upgrade Solr's Lucene dependency from 9.5.0 to the latest
> 9.6.0 version.
> new or existing contributors interested in this ticket:
> * please take a look at the documentation to see what's involved:
> [https://github.com/apache/solr/blob/main/dev-docs/lucene-upgrade.md]
> * optional: add a short note on this ticket to share your intention to work
> on this
> * make the upgrade changes (and of course do feel free to ask any questions
> that may arise)
> * open a pull request
> Solr committers, please could you:
> * help review/merge/backport etc. the pull request for this ticket
> * try and leave the upgrade itself as an opportunity for someone else to
> also contribute to the project
> Thanks everyone!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@solr.apache.org
For additional commands, e-mail: issues-h...@solr.apache.org
