Track and report unpack performance #3610
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
dbutenhof
added
Server
Code Infrastructure
Audit
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
This is a thought experiment, based off our earlier discussion of tracking an
asynchronous unpacker. I thought it would be fun to see some information on
tarball compression ratios, as well as tracking the min/max unpack time in
accessible metadata rather than just in logs. I need to jog back to Horreum,
but before diving back into the pool of Java, I took a recreational break...
So I added a simple `server.unpack-perf` metadata, which is a JSON block like
`{"min": <seconds>, "max": <seconds>, "count": <unpack_count>}`, and then
played with the report generator to get some statistics. The sample below is
for a `runlocal`, with a few small-ish tarballs. The big catch in deploying
this would be that none of the existing datasets will have
`server.unpack-perf` until they're unpacked again, which definitely reduces
the usefulness of this thought experiment.
Nevertheless, I figured I might as well post it for consideration.
```
Cache report:
5 datasets currently unpacked, consuming 51.7 MB
8 datasets have been unpacked a total of 15 times
The least recently used cache was referenced today, pbench-user-benchmark_example-vmstat_2018.10.24T14.38.18
The most recently used cache was referenced today, uperf_rhel8.1_4.18.0-107.el8_snap4_25gb_virt_2019.06.21T01.28.57
The smallest cache is 4.1 kB, nometadata
The biggest cache is 19.6 MB, trafficgen_basic-forwarding-example_tg:trex-profile_pf:forwarding_test.json_ml:5_tt:bs__2019-08-27T14:58:38
The worst compression ratio is 22.156%, uperf_rhel8.1_4.18.0-107.el8_snap4_25gb_virt_2019.06.21T01.28.57
The best compression ratio is 96.834%, pbench-user-benchmark_example-vmstat_2018.10.24T14.38.18
The fastest cache unpack is 0.013 seconds, nometadata
The slowest cache unpack is 0.078 seconds, trafficgen_basic-forwarding-example_tg:trex-profile_pf:forwarding_test.json_ml:5_tt:bs__2019-08-27T14:58:38
The fastest cache unpack streaming rate is 253.666 Mb/second, trafficgen_basic-forwarding-example_tg:trex-profile_pf:forwarding_test.json_ml:5_tt:bs__2019-08-27T14:58:38
The slowest cache unpack streaming rate is 0.133 Mb/second, nometadata
```
The `pbench-tree-manage` utility supports a deep `ARCHIVE` tree display and also managed the periodic background cache reclamation. Curious after we saw a failed reclaim, I wanted to play with it a bit and realized that it does a full (`search`) discovery unconditionally. First, even with `--display` (which now that we have a proper report generator is rarely necessary) we probably can use the faster SQL discovery most of the time, although I added an option to select the slower `--search` discovery. More importantly, though, the cache reclaimer doesn't need a fully discovered cache manager since it takes the short-cut of examining the `/srv/pbench/cache` tree directly: so we can move the discovery into the `--display` path.
Also a few minor corrections identified during ops review.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
I thought this morning about adding a CLI audit tool to query the audit log. While I wasn't quite motivated enough to write it, it occurred to me to at least cobble up a simple set of audit log statistics while eating breakfast. So here 'tis.
|
Well isn't that cute: an expired SSL cert trying to copy the IT CA cert!
And ... I can't log in to Jenkins to restart the build (just in case), because it seems to just ignore the login. Which might conceivably be related ... I suppose it's telling me to "enjoy my weekend and get off the computer"... |
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
Some cleanup and review comments.
webbnh
previously approved these changes
Mar 5, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I added a simple
server.unpack-perfmetadata, which is a JSON block like{"min": <seconds>, "max": <seconds>, "count": <unpack_count>}, and then played with the report generator to get some statistics.I also wrote a report of the
Audittable contents to summarize the operations, statuses, and users involved in the Pbench Server.The sample below is for a
runlocal, with a few small-ish tarballs. The big catch in deploying this would be that none of the existing datasets will haveserver.unpack-perfuntil they're unpacked again, which somewhat reduces the value of the statistics until they get unpacked again (e.g., for TOC or visualize).Nevertheless, I figured I might as well post it for consideration. Some of the statistics (and how they're calculated and/or represented) are no doubt arguable; but I enjoyed seeing the numbers anyway. 😆