http2: use session not socket timeout, tests by apapirovski · Pull Request #15188 · nodejs/node

apapirovski · 2017-09-04T17:09:46Z

This fixes#15158. Two changes included:

there's no longer a default server socket timeout, it's now replaced by a default session timeout
_unrefActive was being called on the handler rather than the actual session

Marked as WIP as I think we need to investigate whether socket timeout is necessary anywhere and also because this needs a lot more tests, ideally.

Checklist

make -j4 test (UNIX), or vcbuild test (Windows) passes
tests and/or benchmarks are included
commit message follows commit guidelines

Affected core subsystem(s)

http2, test

apapirovski · 2017-09-05T02:18:31Z

Also something I just noticed, this earlier PR should probably be revisited in light of this #15106 — I'm not sure if all of it is relevant anymore?

apapirovski · 2017-09-05T19:45:06Z

@jasnell, @mcollina this should be ready to review. Ideally we would have more tests around this behaviour but it does fix the bug and it does work in a way that makes sense.

mcollina · 2017-09-06T10:07:23Z

CI: https://ci.nodejs.org/job/node-test-pull-request/9959/

mcollina

Can you please check if this affect performance anyhow?

Assuming it does not cause a regression.. LGTM

apapirovski · 2017-09-06T11:52:35Z

Will test later today but I would be surprised if it doesn't affect performance at least a little bit. For one, _unrefActive was being called on something that wasn't a timer and two, the socket timeout was never _unrefActive either.

mcollina · 2017-09-06T13:22:19Z

I think the failures on Mac OS X are unrelated.

Let us know about the performance impact.

@jasnell what do you think?

apapirovski · 2017-09-06T14:00:24Z

Will need to figure out how to get h2load built on a Mac so this might take me a little while. Will update when I make it work.

apapirovski · 2017-09-06T17:31:55Z

@mcollina It's not too bad after all.

 http2/headers.js nheaders=0 n=1000 -1.62 % *** 7.214983e-05
http2/headers.js nheaders=10 n=1000 -1.29 % 6.246195e-02
http2/headers.js nheaders=100 n=1000 -0.90 % 2.357148e-01
http2/headers.js nheaders=1000 n=1000 0.70 % 4.645484e-01

Not too surprising that there's a tiny bit of a hit since before this PR the timeouts were completely bugged and weren't doing anything (basically no-op). Same benchmark with server.timeout = 0; yields this:

 http2/headers.js nheaders=0 n=1000 2.41 % 0.1543180
http2/headers.js nheaders=10 n=1000 1.42 % 0.3769546
http2/headers.js nheaders=100 n=1000 0.82 % 0.5944166
http2/headers.js nheaders=1000 n=1000 0.19 % 0.9174064

Edit: This is after 40 runs on each node version btw.

jasnell · 2017-09-06T20:45:09Z

should be faster to use the kState references here instead of the getters

Posting an updated version shortly. Thanks for the review!

jasnell

LGTM with green CI

jasnell · 2017-09-07T05:19:32Z

There was too much red in the last CI run, trying again: https://ci.nodejs.org/job/node-test-commit/12214/

jasnell · 2017-09-07T05:31:54Z

CI is definitely not green on this one: https://ci.nodejs.org/job/node-test-commit-linuxone/8427/nodes=rhel72-s390x/console

@apapirovski ... can you take a look?

btw, I really appreciate that you've jumped in on these. PR's from new contributors make me super happy :-)

apapirovski · 2017-09-07T11:17:20Z

I think the test might be flaky because of the low timeout. Will look into it.

apapirovski · 2017-09-07T11:39:52Z

@jasnell can we run the CI again? I adjusted the timing to be less strict to match some of the other timeout tests. At 20ms the margin for error was too small.

mcollina · 2017-09-07T16:12:36Z

CI: https://ci.nodejs.org/job/node-test-pull-request/9981/

apapirovski · 2017-09-07T17:24:54Z

Worked this time. Nice! Thanks for the reviews @mcollina & @jasnell.

mcollina · 2017-09-07T17:53:04Z

This does not land cleanly on master, can you please rebase and squash?

Change default timeout to be tracked on the session instead of the socket, as nghttp2 manages the socket and we would need to maintain two sets of timeouts for similar purpose. Also fixes session setTimeout to work as it wasn't getting _unrefActive correctly (was called on the handle). Fixes: nodejs#15158

apapirovski · 2017-09-07T18:04:15Z

All done.

mcollina · 2017-09-07T18:39:18Z

Landed as 46133b5

Change default timeout to be tracked on the session instead of the socket, as nghttp2 manages the socket and we would need to maintain two sets of timeouts for similar purpose. Also fixes session setTimeout to work as it wasn't getting _unrefActive correctly (was called on the handle). Fixes: #15158 PR-URL: #15188 Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

Change default timeout to be tracked on the session instead of the socket, as nghttp2 manages the socket and we would need to maintain two sets of timeouts for similar purpose. Also fixes session setTimeout to work as it wasn't getting _unrefActive correctly (was called on the handle). Fixes: nodejs#15158 PR-URL: nodejs#15188 Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

nodejs-github-bot added the http2 Issues or PRs related to the http2 subsystem. label Sep 4, 2017

apapirovski force-pushed the patch-http2-session-timeout branch 2 times, most recently from fec3fdd to d669eebCompare September 5, 2017 19:44

apapirovski changed the title ~~[WIP] http2: use session not socket timeout, tests~~http2: use session not socket timeout, testsSep 5, 2017

mcollina approved these changes Sep 6, 2017
View reviewed changes

mcollina requested a review from jasnell September 6, 2017 13:22

jasnell reviewed Sep 6, 2017
View reviewed changes

jasnell approved these changes Sep 6, 2017
View reviewed changes

apapirovski force-pushed the patch-http2-session-timeout branch from d669eeb to 5b74d59Compare September 6, 2017 21:03

apapirovski force-pushed the patch-http2-session-timeout branch from 991825b to 16027d2Compare September 7, 2017 11:43

apapirovski force-pushed the patch-http2-session-timeout branch from 16027d2 to e06e885Compare September 7, 2017 18:03

mcollina closed this Sep 7, 2017

apapirovski deleted the patch-http2-session-timeout branch September 7, 2017 20:06

MylesBorins mentioned this pull request Sep 10, 2017
v8.5.0 proposal #15308
Merged

mcollina mentioned this pull request Sep 11, 2017
test: fix flaky test-http2-session-timeout #15328
Closed
2 tasks

MylesBorins added the dont-land-on-v6.x label Sep 20, 2017

Uh oh!

Uh oh!

Conversation

apapirovski commented Sep 4, 2017

Checklist

Affected core subsystem(s)

Uh oh!

apapirovski commented Sep 5, 2017

Uh oh!

apapirovski commented Sep 5, 2017

Uh oh!

mcollina commented Sep 6, 2017

Uh oh!

mcollina left a comment

Choose a reason for hiding this comment

Uh oh!

apapirovski commented Sep 6, 2017

Uh oh!

mcollina commented Sep 6, 2017

Uh oh!

apapirovski commented Sep 6, 2017

Uh oh!

apapirovski commented Sep 6, 2017• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jasnellSep 6, 2017

Choose a reason for hiding this comment

Uh oh!

apapirovskiSep 6, 2017

Choose a reason for hiding this comment

Uh oh!

jasnell left a comment

Choose a reason for hiding this comment

Uh oh!

jasnell commented Sep 7, 2017

Uh oh!

jasnell commented Sep 7, 2017

Uh oh!

apapirovski commented Sep 7, 2017

Uh oh!

apapirovski commented Sep 7, 2017• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mcollina commented Sep 7, 2017

Uh oh!

apapirovski commented Sep 7, 2017

Uh oh!

mcollina commented Sep 7, 2017

Uh oh!

apapirovski commented Sep 7, 2017

Uh oh!

mcollina commented Sep 7, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

apapirovski commented Sep 6, 2017•
edited
Loading

apapirovski commented Sep 7, 2017•
edited
Loading