test-https-set-timeout-server on freebsd

[mhdawson]

• Version: master
• Platform: freebsd10-64
• Subsystem:https

Failure seen when running test against a PR changing a different test:

https://ci.nodejs.org/job/node-test-commit-freebsd/5673/nodes=freebsd10-64/

not ok 614 parallel/test-https-set-timeout-server
  ---
  duration_ms: 1.645
  severity: fail
  stack: |-
    # serverTimeout
    # serverRequestTimeout
    # serverResponseTimeout
    # serverRequestNotTimeoutAfterEnd
    # serverResponseTimeoutWithPipeline
    # idleTimeout
    ok
    Mismatched <anonymous> function calls. Expected 1, actual 0.
        at idleTimeout (/usr/home/iojs/build/workspace/node-test-commit-freebsd/nodes/freebsd10-64/test/parallel/test-https-set-timeout-server.js:149:42)
        at run (/usr/home/iojs/build/workspace/node-test-commit-freebsd/nodes/freebsd10-64/test/parallel/test-https-set-timeout-server.js:31:5)
        at Server.<anonymous> (/usr/home/iojs/build/workspace/node-test-commit-freebsd/nodes/freebsd10-64/test/parallel/test-https-set-timeout-server.js:131:5)
        at emitOne (events.js:96:13)
        at Server.emit (events.js:188:7)
        at TLSSocket.<anonymous> (_http_server.js:314:30)
        at emitNone (events.js:86:13)
        at TLSSocket.emit (events.js:185:7)
        at TLSSocket.Socket._onTimeout (net.js:342:8)
  ...

[mhdawson]

Looking at history don't see any recent changes to the test.

[mhdawson]

Failed again here: https://ci.nodejs.org/job/node-test-commit-freebsd/5675/nodes=freebsd10-64/console

[Trott]

Giving it a quick look, moving to sequential will probably make it more reliable. /cc @nodejs/testing

[Trott]

Failed again today. https://ci.nodejs.org/job/node-test-commit-freebsd/8596/nodes=freebsd10-64/console

not ok 664 parallel/test-https-set-timeout-server
  ---
  duration_ms: 1.263
  severity: fail
  stack: |-
    # serverTimeout
    # serverRequestTimeout
    # serverResponseTimeout
    # serverRequestNotTimeoutAfterEnd
    # serverResponseTimeoutWithPipeline
    # idleTimeout
    ok
    Mismatched <anonymous> function calls. Expected 1, actual 0.
        at idleTimeout (/usr/home/iojs/build/workspace/node-test-commit-freebsd/nodes/freebsd10-64/test/parallel/test-https-set-timeout-server.js:170:44)
        at run (/usr/home/iojs/build/workspace/node-test-commit-freebsd/nodes/freebsd10-64/test/parallel/test-https-set-timeout-server.js:52:5)
        at Server.<anonymous> (/usr/home/iojs/build/workspace/node-test-commit-freebsd/nodes/freebsd10-64/test/parallel/test-https-set-timeout-server.js:152:5)
        at emitOne (events.js:115:13)
        at Server.emit (events.js:210:7)
        at TLSSocket.socketOnTimeout (_http_server.js:385:35)
        at emitNone (events.js:105:13)
        at TLSSocket.emit (events.js:207:7)
        at TLSSocket.Socket._onTimeout (net.js:374:8)

[Trott]

I'm going to run some stress tests on it in parallel vs. sequential and see what happens.

[Trott]

CI stress test on freebsd10-64 with -j 4 --repeat=4:

https://ci.nodejs.org/job/node-stress-single-test/1175/nodes=freebsd10-64/console

[Trott]

Ci stress test on freebsd10-64 with -j 8 --repeat=8:

https://ci.nodejs.org/job/node-stress-single-test/1176/nodes=freebsd10-64/console

[Trott]

8 seems to be enough to cause failures identical to what we're seeing in CI from time to time, so that's good. But the errors are infrequent. (I got 1 in 500 runs.) Trying 16 to see if we can get failures to happen with greater frequency:

https://ci.nodejs.org/job/node-stress-single-test/1177/nodes=freebsd10-64/console

That gives me 92 failures out of 100 runs. Cool. Solution is certainly to move it to sequential but I'll run a stress test on that anyway.

[Trott]

Actually, I wonder if moving the setting of the server timeout to the callback might be the solution...