Investigate test-http2-createwritereq failure on Linux and AIX #17840

Trott · 2017-12-23T15:44:26Z

Version: 10.0.0-pre
Platform: aix61-ppc64
Subsystem: test http2

https://ci.nodejs.org/job/node-test-commit-aix/11376/nodes=aix61-ppc64/console

not ok 934 parallel/test-http2-createwritereq
  ---
  duration_ms: 1.818
  severity: crashed
  stack: |-
    oh no!
    exit code: CRASHED (Signal: 11)

Trott · 2017-12-23T15:44:39Z

I guess I'll start by pinging the test author: @apapirovski

Trott · 2017-12-23T15:47:00Z

Signal 11 = segmentation fault. I'm guessing AIX on CI is not configured to preserve core files, but let's ask. Ping @nodejs/build.

Trott · 2017-12-23T15:47:25Z

And ping @nodejs/platform-aix for additional troubleshooting and/or suggestions on how to proceed.

addaleax · 2017-12-23T17:01:26Z

Is there any way to run stress tests at a given git SHA or something like that? #17406 or #17718 would seem like possible causes to me

Trott · 2017-12-23T17:06:18Z

Is there any way to run stress tests at a given git SHA or something like that? #17406 or #17718 would seem like possible causes to me

In the past, I've created branches at specific SHAs, pushed them to my fork (possibly to my master branch, I can't remember if I got it to work off another branch or not), and run the stress test off of my fork.

Trott · 2017-12-23T17:12:23Z

I guess it might make sense to run a stress test against master to see if this is reproducible.

Running it serially: https://ci.nodejs.org/job/node-stress-single-test/1574/nodes=aix61-ppc64/

Running it in parallel (96 processes): https://ci.nodejs.org/job/node-stress-single-test/1575/nodes=aix61-ppc64/

gireeshpunathil · 2017-12-24T05:06:03Z

Ran a 2000 times locally with no luck on recreate. Either wait for @apapirovski to seek hints on possible causes, or gain login access to CI. /cc @mhdawson @gibfahn

The test does not seem to use too much of memory (few bytes of data transported through http2 and validate it passes through properly with differing encodings) which rules out native memory constraints as a source of trouble (which was a known difference between CI and local AIX boxes that led to some failures.)

Trott · 2017-12-24T05:18:32Z

Reproduced it in CI stress tests, so at least it's reproducible in that environment. Running serially, it got 31 failures in 9999 runs:

9999   OK: 9968   NOT OK: 31   TOTAL: 9999
+ '[' 31 '!=' 0 ']'
+ echo The test is flaky.
The test is flaky.
+ exit 1
Build step 'Conditional steps (multiple)' marked build as failure
Notifying upstream projects of job completion
Finished: FAILURE

...so running under load is not a precondition for failure.

apapirovski · 2017-12-24T12:52:15Z

I don't think that test does anything particularly weird. If I had to guess, other http2 tests that write stuff might be flaky on that platform too. It's reasonably likely that the recent changes to http2 are responsible. I'll try to have a look when I have time but as @addaleax mentioned, a good place to start is running a couple of stress tests against master rewound to the two PRs above.

gireeshpunathil · 2017-12-24T14:19:42Z

thanks @apapirovski . @Trott - is it (artificially rewinding the two PRs and running stress on it) something which can be done through the existing build jobs or custom scripts?

In terms of isolation: both the PRs contain considerable amount of code changes that makes it difficult for me to manually locate the crash reason. On the other hand, if we isolate the issue into one of these PRs, still the failure reason needs to come bottom up (starting with the crashing context). So I prefer (in my way of problem determination) performing core dump analysis directly in the CI machine.

Trott · 2017-12-24T16:48:02Z

❌ = failure, ✅ = success

❌ Stress test rewound to e0e6b68 (right after #11718 landed): https://ci.nodejs.org/job/node-stress-single-test/1576/nodes=aix61-ppc64/

✅ Stress test rewound to e3b44b6 (one commit before the above so that it's just the first of the two commits that landed in #11718): https://ci.nodejs.org/job/node-stress-single-test/1577/nodes=aix61-ppc64/

✅ Stress test rewound to e554bc8 (right before #11718): https://ci.nodejs.org/job/node-stress-single-test/1579/nodes=aix61-ppc64/

Trott · 2017-12-24T19:23:25Z

@gireeshpunathil Stress tests aren't done running yet, but it sure looks like the issue is probably with e0e6b68. (Stress test at that commit failed, and the stress test at each of the two commits before it are both clean so far....

Trott · 2017-12-25T02:06:39Z

Stress tests are done and the evidence does point to e0e6b68....

gireeshpunathil · 2017-12-25T07:37:02Z

thanks @Trott , for the info

addaleax · 2017-12-25T10:52:36Z

Btw, I could reproduce this locally on Linux, so it isn’t AIX-specific … it’s my patch that caused this so I’ll try to figure out what’s going on here :)

gireeshpunathil · 2017-12-25T11:58:16Z

thanks @addaleax , that is good news as in we don't need to access CI. How frequent is the crash? similar to that of AIX?

addaleax · 2017-12-25T12:09:30Z

It’s less than 1 in 1000 – I think that’s about the same rate.

From some initial poking around, it seems like one of the Http2Session instances is garbage collected while it is handling data from the socket (i.e. inside nghttp2_session_mem_recv()).

Keep a local handle as a reference to the JS `Http2Session` object so that it will not be garbage collected when inside an `Http2Scope`, because the presence of the latter usually indicates that further actions on the session object are expected. Strictly speaking, storing the `session_handle_` as a property on the scope object is not necessary, but this is not very costly and makes the code more obviously correct. Fixes: nodejs#17840

addaleax · 2017-12-25T19:23:00Z

Reproducible test case + fix @ #17863

removing the test label here because it’s not an issue with the test itself

Keep a local handle as a reference to the JS `Http2Session` object so that it will not be garbage collected when inside an `Http2Scope`, because the presence of the latter usually indicates that further actions on the session object are expected. Strictly speaking, storing the `session_handle_` as a property on the scope object is not necessary, but this is not very costly and makes the code more obviously correct. Fixes: #17840 PR-URL: #17863 Fixes: #17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

PR-URL: nodejs#17863 Fixes: nodejs#17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Keep a local handle as a reference to the JS `Http2Session` object so that it will not be garbage collected when inside an `Http2Scope`, because the presence of the latter usually indicates that further actions on the session object are expected. Strictly speaking, storing the `session_handle_` as a property on the scope object is not necessary, but this is not very costly and makes the code more obviously correct. Fixes: nodejs#17840 PR-URL: nodejs#17863 Fixes: nodejs#17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Backport-PR-URL: #18050 PR-URL: #17863 Fixes: #17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Keep a local handle as a reference to the JS `Http2Session` object so that it will not be garbage collected when inside an `Http2Scope`, because the presence of the latter usually indicates that further actions on the session object are expected. Strictly speaking, storing the `session_handle_` as a property on the scope object is not necessary, but this is not very costly and makes the code more obviously correct. Fixes: #17840 Backport-PR-URL: #18050 PR-URL: #17863 Fixes: #17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Backport-PR-URL: #18050 PR-URL: #17863 Fixes: #17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Keep a local handle as a reference to the JS `Http2Session` object so that it will not be garbage collected when inside an `Http2Scope`, because the presence of the latter usually indicates that further actions on the session object are expected. Strictly speaking, storing the `session_handle_` as a property on the scope object is not necessary, but this is not very costly and makes the code more obviously correct. Fixes: #17840 Backport-PR-URL: #18050 PR-URL: #17863 Fixes: #17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Backport-PR-URL: nodejs#18050 PR-URL: nodejs#17863 Fixes: nodejs#17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Keep a local handle as a reference to the JS `Http2Session` object so that it will not be garbage collected when inside an `Http2Scope`, because the presence of the latter usually indicates that further actions on the session object are expected. Strictly speaking, storing the `session_handle_` as a property on the scope object is not necessary, but this is not very costly and makes the code more obviously correct. Fixes: nodejs#17840 Backport-PR-URL: nodejs#18050 PR-URL: nodejs#17863 Fixes: nodejs#17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Backport-PR-URL: #18050 Backport-PR-URL: #20456 PR-URL: #17863 Fixes: #17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Keep a local handle as a reference to the JS `Http2Session` object so that it will not be garbage collected when inside an `Http2Scope`, because the presence of the latter usually indicates that further actions on the session object are expected. Strictly speaking, storing the `session_handle_` as a property on the scope object is not necessary, but this is not very costly and makes the code more obviously correct. Fixes: #17840 Backport-PR-URL: #18050 Backport-PR-URL: #20456 PR-URL: #17863 Fixes: #17840 Reviewed-By: Anatoli Papirovski <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

Trott added aix Issues and PRs related to the AIX platform. http2 Issues or PRs related to the http2 subsystem. test Issues and PRs related to the tests. labels Dec 23, 2017

maclover7 added the flaky-test Issues and PRs related to the tests with unstable failures on the CI. label Dec 25, 2017

addaleax removed the aix Issues and PRs related to the AIX platform. label Dec 25, 2017

Trott changed the title ~~Investigate test-http2-createwritereq failure on AIX~~ Investigate test-http2-createwritereq failure on Linux and AIX Dec 25, 2017

addaleax mentioned this issue Dec 25, 2017

http2: keep session objects alive during Http2Scope #17863

Closed

3 tasks

addaleax added flaky-test Issues and PRs related to the tests with unstable failures on the CI. and removed flaky-test Issues and PRs related to the tests with unstable failures on the CI. test Issues and PRs related to the tests. labels Dec 25, 2017

jasnell closed this as completed in 57594d2 Dec 28, 2017

addaleax mentioned this issue Feb 13, 2018

src: prevent persistent handle resource leaks #18656

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Investigate test-http2-createwritereq failure on Linux and AIX #17840

Investigate test-http2-createwritereq failure on Linux and AIX #17840

Trott commented Dec 23, 2017

Trott commented Dec 23, 2017

Trott commented Dec 23, 2017

Trott commented Dec 23, 2017

addaleax commented Dec 23, 2017

Trott commented Dec 23, 2017

Trott commented Dec 23, 2017

gireeshpunathil commented Dec 24, 2017

Trott commented Dec 24, 2017 •

edited

Loading

apapirovski commented Dec 24, 2017

gireeshpunathil commented Dec 24, 2017

Trott commented Dec 24, 2017 •

edited

Loading

Trott commented Dec 24, 2017

Trott commented Dec 25, 2017

gireeshpunathil commented Dec 25, 2017

addaleax commented Dec 25, 2017

gireeshpunathil commented Dec 25, 2017

addaleax commented Dec 25, 2017

addaleax commented Dec 25, 2017 •

edited

Loading

Investigate test-http2-createwritereq failure on Linux and AIX #17840

Investigate test-http2-createwritereq failure on Linux and AIX #17840

Comments

Trott commented Dec 23, 2017

Trott commented Dec 23, 2017

Trott commented Dec 23, 2017

Trott commented Dec 23, 2017

addaleax commented Dec 23, 2017

Trott commented Dec 23, 2017

Trott commented Dec 23, 2017

gireeshpunathil commented Dec 24, 2017

Trott commented Dec 24, 2017 • edited Loading

apapirovski commented Dec 24, 2017

gireeshpunathil commented Dec 24, 2017

Trott commented Dec 24, 2017 • edited Loading

Trott commented Dec 24, 2017

Trott commented Dec 25, 2017

gireeshpunathil commented Dec 25, 2017

addaleax commented Dec 25, 2017

gireeshpunathil commented Dec 25, 2017

addaleax commented Dec 25, 2017

addaleax commented Dec 25, 2017 • edited Loading

Trott commented Dec 24, 2017 •

edited

Loading

Trott commented Dec 24, 2017 •

edited

Loading

addaleax commented Dec 25, 2017 •

edited

Loading