Improve Geronimo Parallelism#202
Conversation
mmusgrov
left a comment
mmusgrov
left a comment
There was a problem hiding this comment.
The PR is still running but the changes look good (I agree that we should strive to get the best results possible for other TMs so that we get a true comparison).
I don't know what exactly the checksumEnabled flag does (the geronimo javadoc for it doesn't say) but the ApacheTomEE config seems to set it to true by default. I wonder what the impact of not setting it is.
| 10,// "threadsWaitingForceThreshold"}); | ||
| xidFactory, | ||
| new File(buildDir) | ||
| "org.objectweb.howl.log.BlockLogBuffer", |
There was a problem hiding this comment.
Thanks @jmfinelli for your PR, as I am not familiar with Geronimo configuration it might be useful to keep the information from the comments we had?
There was a problem hiding this comment.
org.apache.geronimo.transaction.log.HOWLLog already reports those variable names, that's why I removed them
I should have opened a Jira to explain why I raised this PR. Sorry about that. Apparently, we're experiencing a strange failure in the PERFORMANCE job related to Geronimo: its transaction logs gets corrupted and the job gets stuck. I'm trying to fix this issue by modifying the HOWL parameters. So far, no luck. I suppose that the limited resources of the runner are the problem...but we cannot do much about that. |
|
jmfinelli
force-pushed
the
improve_geronimo_parallelism
branch
from
efbabd1 to
6dddfcc
Compare
jmfinelli
force-pushed
the
improve_geronimo_parallelism
branch
from
c7d0720 to
8d39465
Compare
|
I'm going to re-run all GH actions in this PR but I would like to highlight that a full passed was already achieved |
jmfinelli
force-pushed
the
improve_geronimo_parallelism
branch
from
151feca to
89e716f
Compare
marcosgopen
left a comment
marcosgopen
left a comment
There was a problem hiding this comment.
LGTM, thanks @jmfinelli !
| name: Testing PERFORMANCE with JDK ${{ inputs.jdk_version }} | ||
| # --------------------------------------------------------------------------- | ||
| # Build Narayana once and determine which benchmarks to run | ||
| # --------------------------------------------------------------------------- |
There was a problem hiding this comment.
Thanks Marco! I will squash all commits together.
As a note, the performance benchmark with 1600 threads can still fail because of Geronimo but the probability is definitely smaller than before. cc @mmusgrov @tomjenkinson
jmfinelli
force-pushed
the
improve_geronimo_parallelism
branch
from
c353691 to
77399ec
Compare
3 participants
Trying to address #203