Re: [PR] Enable async profiler [flink-benchmarks]
Zakelly merged PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
Zakelly commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2151314927 > @Zakelly @pnowojski do you need anything else from me to progress this? Would one of you be able to approve this PR to run through the workflow so I can verify its good to go? Sry for the late reply. I tried the newest commit and the good news is, the benchmark duration remains same with previous setup (JMH 1.19), and turning the profiler on or off does not affect the duration. I still need some time to debug jenkins script and verify there is no regression with previous results before putting this in daily run, but this PR LGTM. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2151090653 @Zakelly @pnowojski do you need anything else from me to progress this? Would one of you be able to approve this PR to run through the workflow so I can verify its good to go? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
Zakelly commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2139458543 > @Zakelly I've pushed [decdc8e](https://github.com/apache/flink-benchmarks/commit/decdc8e2d161584a798461b4f381adba7cee731f) which restores the JMH defaults from 1.19. Hopefully that addresses the issues with the runtime under Jenkins. I'd just as happily revert that commit but it de-couples this PR from reasoning about the implications for changing the warmup and measurement times. I see. Thanks for your investigation. I'll give it a try. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2138664126 @Zakelly I've pushed https://github.com/apache/flink-benchmarks/pull/90/commits/decdc8e2d161584a798461b4f381adba7cee731f which restores the JMH defaults from 1.19. Hopefully that addresses the issues with the runtime under Jenkins. I'd just as happily revert that commit but it de-couples this PR from reasoning about the implications for changing the warmup and measurement times. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2138632929 I've confirmed the runtime duration increase comes from upgrading JMH from 1.19 to 1.21. The minimum version for async profiler looks to be JMH 1.31. So if we want flame graphs 1.37 still feels like the right choice. What I haven't been able to determine is what the change in warmup behaviour implies for the insights of the current benchmark suite and how comparable through time the results would be. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on PR #90:
URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2138478974
> It compiles & install flink before run benchmarks via
/mnt/jenkins/tools/hudson.tasks.Maven_MavenInstallation/M3/bin/mvn
-Dflink.version=1.20-SNAPSHOT clean install exec:exec -Penable-async-profiler
-DskipTests -Drat.skip=true '-Dbenchmarks=org.apache.flink.state.benchmark.*'
'-DbenchmarkExcludes=org.apache.flink.benchmark.full.*'
-DexecutableJava=/usr/lib/jvm/java-1.8.0-openjdk-1.8.0.372.b07-4.0.2.al8.x86_64/bin/java
-DasyncProfilerLib=/opt/async-profiler-3.0-linux-x64/lib/libasyncProfiler.so
-DprofResultPath=./profile-result
>It would be great if you could provide some options to speed this up.
It looks to me like it was handling the excludes and includes correctly
which is what I was concerned about. I can't reproduce a slower runtime with
the profiler on
With profiler
```
mvn -Dflink.version=1.20-SNAPSHOT -DexecutableJava=${JAVA_HOME}/bin/java
-DasyncProfilerLib=/home/sam/src/async-profiler/async-profiler-3.0-linux-x64/lib/libasyncProfiler.so
-Dbenchmarks="SerializationFrameworkMiniBenchmarks.serializerHeavyString"
exec:exec --offline
...
snip
...
# JMH version: 1.37
# VM version: JDK 11.0.23, OpenJDK 64-Bit Server VM, 11.0.23+9
# VM invoker: /home/sam/.sdkman/candidates/java/11.0.23-tem/bin/java
# VM options: -Djava.rmi.server.hostname=127.0.0.1
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=false -Dcom.sun.management.jmxremote.ssl
# Blackhole mode: full + dont-inline hint (auto-detected, use
-Djmh.blackhole.autoDetect=false to disable)
# Warmup: 10 iterations, 10 s each
# Measurement: 10 iterations, 10 s each
# Timeout: 10 min per iteration
# Threads: 1 thread, will synchronize iterations
# Benchmark mode: Throughput, ops/time
# Benchmark:
org.apache.flink.benchmark.SerializationFrameworkMiniBenchmarks.serializerHeavyString
# Run progress: 0.00% complete, ETA 00:10:00
...
snip
...
[INFO]
[INFO] BUILD SUCCESS
[INFO]
[INFO] Total time: 11:17 min
[INFO] Finished at: 2024-05-30T12:10:42+12:00
[INFO]
```
without profiler
```
mvn -Dflink.version=1.20-SNAPSHOT -DexecutableJava=${JAVA_HOME}/bin/java
-Dbenchmarks="SerializationFrameworkMiniBenchmarks.serializerHeavyString"
exec:exec --offline
...
snip
...
# JMH version: 1.37
# VM version: JDK 11.0.23, OpenJDK 64-Bit Server VM, 11.0.23+9
# VM invoker: /home/sam/.sdkman/candidates/java/11.0.23-tem/bin/java
# VM options: -Djava.rmi.server.hostname=127.0.0.1
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=false -Dcom.sun.management.jmxremote.ssl
# Blackhole mode: full + dont-inline hint (auto-detected, use
-Djmh.blackhole.autoDetect=false to disable)
# Warmup: 10 iterations, 10 s each
# Measurement: 10 iterations, 10 s each
# Timeout: 10 min per iteration
# Threads: 1 thread, will synchronize iterations
# Benchmark mode: Throughput, ops/time
# Benchmark:
org.apache.flink.benchmark.SerializationFrameworkMiniBenchmarks.serializerHeavyString
# Run progress: 0.00% complete, ETA 00:10:00
...
snip
...
[INFO]
[INFO] BUILD SUCCESS
[INFO]
[INFO] Total time: 11:18 min
[INFO] Finished at: 2024-05-30T12:27:06+12:00
[INFO]
```
~So I'm not really sure what to suggest.~
Wait a minute. This PR changes the version of JMH. Its not the profiler
being ON or OFF that affects the runtime its the JMH version.
going back to main I get
```
# JMH version: 1.19
# VM version: JDK 11.0.23, VM 11.0.23+9
# VM invoker: /home/sam/.sdkman/candidates/java/11.0.23-tem/bin/java
# VM options: -Djava.rmi.server.hostname=127.0.0.1
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=false -Dcom.sun.management.jmxremote.ssl
# Warmup: 10 iterations, 1 s each
# Measurement: 10 iterations, 1 s each
# Timeout: 10 min per iteration
# Threads: 1 thread, will synchronize iterations
# Benchmark mode: Throughput, ops/time
# Benchmark:
org.apache.flink.benchmark.SerializationFrameworkMiniBenchmarks.serializerHeavyString
# Run progress: 0.00% complete, ETA 00:01:00
```
This may be related to changes in warmups in [JMH
1.21](https://mail.openjdk.org/pipermail/jmh-dev/2018-May/002753.html)
> *) Defaults for warmup/measurement/forks were reconsidered
Re: [PR] Enable async profiler [flink-benchmarks]
Zakelly commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2137299531 Here's the full log before the pipeline expired. [jenkins.tar.gz](https://github.com/apache/flink-benchmarks/files/15484257/jenkins.tar.gz) It compiles & install flink before run benchmarks via `/mnt/jenkins/tools/hudson.tasks.Maven_MavenInstallation/M3/bin/mvn -Dflink.version=1.20-SNAPSHOT clean install exec:exec -Penable-async-profiler -DskipTests -Drat.skip=true '-Dbenchmarks=org.apache.flink.state.benchmark.*' '-DbenchmarkExcludes=org.apache.flink.benchmark.full.*' -DexecutableJava=/usr/lib/jvm/java-1.8.0-openjdk-1.8.0.372.b07-4.0.2.al8.x86_64/bin/java -DasyncProfilerLib=/opt/async-profiler-3.0-linux-x64/lib/libasyncProfiler.so -DprofResultPath=./profile-result` It would be great if you could provide some options that could help speed this up. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2137002719 > And I find the benchmark runs extremely slow with the profiler on. The ETA is up to 8 days for full set of our daily run. So I guess this is best only used for single tests that are triggered manually, right? Yikes thats a bit sad. Running a single test (`SerializationFrameworkMiniBenchmarks.serializerHeavyString`) locally I get 10 mins estimate with or without the profiler and a runtime that seems comparable. If I widen the set of benchmarks I get the same ETA with or without the profiler. So I'm slightly suspicious that the excludes or default includes behaviour is being changed but without knowing what command line your seeing that from its hard to know. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on code in PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1618517159 ## benchmark.sh: ## @@ -0,0 +1,54 @@ +#!/usr/bin/env bash Review Comment: resolved with https://github.com/apache/flink-benchmarks/pull/90/commits/74e4f4f56dab07c70e03690279f96a8f168830f4 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
Zakelly commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2136859786 And I find the benchmark runs extremely slow with the profiler on. The ETA is up to 8 days for full set of our daily run. So I guess this is best only used for single tests that are triggered manually, right? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
Zakelly commented on code in PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1618445050 ## benchmark.sh: ## @@ -0,0 +1,54 @@ +#!/usr/bin/env bash Review Comment: `chown +x` to make this executable? ## pom.xml: ## @@ -67,6 +67,7 @@ under the License. org.apache.flink.benchmark.full.*,org.apache.flink.state.benchmark.*,org.apache.flink.scheduler.benchmark.* .* java + Review Comment: How about adding a property for path for profiling result? And I suggest a relative path under current path for default value. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
showuon commented on code in PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1618289766 ## README.md: ## @@ -64,6 +64,20 @@ java -jar target/benchmarks.jar "" -lp java -jar target/benchmarks.jar "org.apache.flink.state.benchmark.*" -p "backendType=ROCKSDB" ``` +## Generating Flame graphs + +JMH has support for enabling profilers when running benchmarks, in particular [async profiler](https://github.com/async-profiler/async-profiler) which can generate flame graphs of the call stack. However, async profiler requires linking with native code which needs downloaded manually which means its location is user and architecture specific. To enable async profiler support run the benchmarks as follows: + +with maven +``` + java -jar target/benchmarks.jar -rf csv "" -DasyncProfilerLib= Review Comment: nit: additional space at the beginning. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2136485697 @Zakelly any news on the jenkins env? I'm largely hoping its a no-op there right now. @pnowojski readme and executable bit set. Is there anything else your looking for? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on code in PR #90:
URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1618157436
##
benchmark.sh:
##
@@ -0,0 +1,48 @@
+#!/usr/bin/env bash
+
+JAVA_ARGS=()
+JMH_ARGS=()
+BINARY="java"
+BENCHMARK_PATTERN=
+
+while getopts ":j:c:b:e:p:a:m:h" opt; do
+ case $opt in
+j) JAVA_ARGS+=("${OPTARG}")
+;;
+c) CLASSPATH_ARG="${OPTARG}"
+;;
+b) BINARY="${OPTARG}"
+;;
+p) PROFILER_ARG="${OPTARG:+-prof ${OPTARG}}"
+# conditional prefixing inspired by
https://stackoverflow.com/a/40771884/1389220
+;;
+a) JMH_ARGS+=("${OPTARG}")
+;;
+e) BENCHMARK_EXCLUDES="${OPTARG:+-e ${OPTARG}}"
+;;
+m) BENCHMARK_PATTERN="${OPTARG}"
+ echo "parsing -m"
+;;
+h)
+ 1>&2 cat << EOF
+usage: TODO
+EOF
+ exit 1
+;;
+\?) echo "Invalid option -$opt ${OPTARG}" >&2
+exit 1
+;;
+ esac
+done
+shift "$(($OPTIND -1))"
Review Comment:
I had a go at doing it in python
https://github.com/SamBarker/flink-benchmarks/blob/generate-flamegraphs-py/run-benchmarks.py
life got quite funky trying to pass `--add-opens` to something that actually
understood long arguments.
I can merge that into this PR if preferred but I'm not sure it actually
improves things here.
The other option might be to look at
[ap-loader](https://github.com/jvm-profiling-tools/ap-loader) which would allow
us to ensure that the native libraries are available.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
pnowojski commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2134747996 > Yes that's almost right. The profile is enabled automatically by specifying the path the profiler shared library (as opposed to running -P enable-async-profiler) Async profiler needs downloaded manually and the path is platform specific I didn't want to tackle automation of that unless there was some interest in the idea. Can you document this process in the README.md? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
Zakelly commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2134617507 Thanks for the contribution! I'll try this out in our `jenkins` env. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2134019437 > AFAIU, by default profiling is not turned on. One has to activate enable-async-profiler profile? Yes that's almost right. The profile is enabled automatically by specifying the path the profiler shared library (as opposed to running `-P enable-async-profiler`) Async profiler needs downloaded manually and the path is platform specific I didn't want to tackle automation of that unless there was some interest in the idea. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on code in PR #90:
URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1616369948
##
pom.xml:
##
@@ -290,25 +291,31 @@ under the License.
${skipTests}
test
-
${executableJava}
+
${basedir}/benchmark.sh
Review Comment:
The interface with maven is identical so how the Java env is controlled
stays consistent. The change is in how Jenkins triggers the benchmarks.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
pnowojski commented on code in PR #90:
URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1616223910
##
benchmark.sh:
##
@@ -0,0 +1,54 @@
+#!/usr/bin/env bash
Review Comment:
I would mark this as executable.
##
benchmark.sh:
##
@@ -0,0 +1,48 @@
+#!/usr/bin/env bash
+
+JAVA_ARGS=()
+JMH_ARGS=()
+BINARY="java"
+BENCHMARK_PATTERN=
+
+while getopts ":j:c:b:e:p:a:m:h" opt; do
+ case $opt in
+j) JAVA_ARGS+=("${OPTARG}")
+;;
+c) CLASSPATH_ARG="${OPTARG}"
+;;
+b) BINARY="${OPTARG}"
+;;
+p) PROFILER_ARG="${OPTARG:+-prof ${OPTARG}}"
+# conditional prefixing inspired by
https://stackoverflow.com/a/40771884/1389220
+;;
+a) JMH_ARGS+=("${OPTARG}")
+;;
+e) BENCHMARK_EXCLUDES="${OPTARG:+-e ${OPTARG}}"
+;;
+m) BENCHMARK_PATTERN="${OPTARG}"
+ echo "parsing -m"
+;;
+h)
+ 1>&2 cat << EOF
+usage: TODO
+EOF
+ exit 1
+;;
+\?) echo "Invalid option -$opt ${OPTARG}" >&2
+exit 1
+;;
+ esac
+done
+shift "$(($OPTIND -1))"
Review Comment:
I would suggest to use python or java wrapper instead of bash, if only for
the sake of being able to use a nice args library. For example take a look at
https://github.com/apache/flink-benchmarks/blob/master/regression_report.py or
some other python script in this repo's root.
##
pom.xml:
##
@@ -290,25 +291,31 @@ under the License.
${skipTests}
test
-
${executableJava}
+
${basedir}/benchmark.sh
Review Comment:
Are the executable commands the same? There are some scripts (in
[jenkins](https://github.com/apache/flink-benchmarks/tree/master/jenkinsfiles)?)
that are relaying for those commands to behave the way they were behaving. If
something changes, they would have to be changed as well.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on code in PR #90:
URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1615688702
##
pom.xml:
##
@@ -54,11 +54,11 @@ under the License.
${java.version}
1.7.36
1.2.17
- 1.19
+ 1.37
4.13.2
1.11.1
3.4.6
- 1.6.0
+ 3.3.0
Review Comment:
The old version had issues trying to log `-p` so went for the most recent
release.
##
pom.xml:
##
@@ -54,11 +54,11 @@ under the License.
${java.version}
1.7.36
1.2.17
- 1.19
+ 1.37
Review Comment:
Needed a newer version for async-profiler support in JMH so went for the
most recent release.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on code in PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1615426480 ## benchmark.sh: ## @@ -0,0 +1,54 @@ +#!/usr/bin/env bash Review Comment: One final call out. This is not currently checked in with the executable bit set as I'm not sure what the projects preference would be for that. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] Enable async profiler [flink-benchmarks]
SamBarker commented on PR #90: URL: https://github.com/apache/flink-benchmarks/pull/90#issuecomment-2132564249 I think the `benchmark-java17` profile could potentially be simplified to specify a property with just the additional JVM_ARGS (possibly changing `-j` to be single arg inline with `-p`). However I wanted to gauge interest in this change before I went any further with potential refactoring. I also think it would be very helpful to capture the flamegraphs as part of the information included in `flink-speed` but I have no idea about the implications of that or the required setup for the build server so haven't perused that option as yet. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
Re: [PR] enable async profiler [flink-benchmarks]
SamBarker commented on code in PR #90:
URL: https://github.com/apache/flink-benchmarks/pull/90#discussion_r1615355106
##
pom.xml:
##
@@ -345,6 +346,8 @@ under the License.
-classpath
org.openjdk.jmh.Main
+
-prof
+
async:libPath=${asyncProfilerLib};output=flamegraph;dir=/tmp/profile-results
Review Comment:
Argh. this fails if `asyncProfilerLib` is invalid.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
