1447411 - Web Replay: Get mochitests working

Status: RESOLVED INVALID

(Core :: General, enhancement)

Description

[Brian Hackett [Laid off!]]

There are several mochitests for Web Replay in place already, which are nice for testing locally but have a number of problems causing unexpected failures.  This bug is about fixing these issues and getting a clean bill of health for the tests from the mochitest harness.

Comment 1

[Brian Hackett [Laid off!]]

Attachment: Part 1 - Listen for browser close and session store flush IPDL messages in the middleman.

The middleman process wasn't shutting down or flushing session store state in response to IPDL messages to close it, leading the mochitest harness to kill it.  This is because the messages were being forwarded to the replaying process, which ignored them.  This patch handles these messages in the middleman.

Comment 2

[Brian Hackett [Laid off!]]

Attachment: Part 2 - Add utility function for suppressing uncaught rejections from current test.

There are some uncaught rejections on the web replay tests due to the client trying to interact with frame actors that no longer exist.  I haven't tracked down the underlying cause here but fixing it will probably require changing the client, which I don't want to do right now.  Anyways, suppressing these rejections is tricky because neither of the two APIs that PromiseTestUtils provides are very suitable.  expectUncaughtRejection does not work because the rejections don't always happen, and whitelistRejectionsGlobally affects the behavior of later tests.  This patch adds a third API, removeUncaughtRejections, that filters out uncaught rejections from the current test which match a pattern.

Comment 3

[Brian Hackett [Laid off!]]

Attachment: Part 3 - Watch for exception when getting environment from dead frame.

Catch an exception that was being thrown when the debugger server tried to use a frame that wasn't live.  There might be an underlying issue here related to the invalid frame actors thing mentioned in comment 2.

Comment 4

[Brian Hackett [Laid off!]]

Attachment: Part 4 - Avoid uncaught rejections in web replay tests.

Avoid all uncaught rejections on web replay tests, by fixing a bug where listeners for one test were being called on future tests, and by suppressing the remaining rejections with the API added in part 2.

Comment 5

[Brian Hackett [Laid off!]]

https://hg.mozilla.org/projects/ash/rev/a16e75460f65d96e47e4bfe124fe0662ab981ae4
https://hg.mozilla.org/projects/ash/rev/04af13369f557a6037ac79f236fd1ee11b4a059e
https://hg.mozilla.org/projects/ash/rev/e169933ad875fde3a740ea0f5e3b579ac3fdb051
https://hg.mozilla.org/projects/ash/rev/60aa707d9ed6f954650ca1fc3c0f214a3c2cf5d1

Comment 6

[Brian Hackett [Laid off!]]

Attachment: Part 5 - Globally whitelist unknown actor rejection for web replay tests.

The function added in part 2 hasn't been working very reliably, so this patch removes it and globally whitelists the rejections on the web replay tests.

https://hg.mozilla.org/projects/ash/rev/b7540bed619c4b8630e1993c61cdf2e488a9984f

Comment 7

[Brian Hackett [Laid off!]]

Attachment: Part 6 - Suppress messages from child process to dead browsers in the UI process.

This patch fixes an issue stemming from Part 1.  When the middleman receives browser close messages it both handles them itself and forwards them to the child process.  If the child process is still recording then it will send responses to those close messages, and since the middleman sent them as well the UI process receives two sets of messages and gets confused.  The close messages from the middleman are the canonical ones, so we just drop any messages coming from the child process for a browser that the UI process has destroyed.

https://hg.mozilla.org/projects/ash/rev/2edfde574e9fe9c07de73ddd9b8525543ca5eb9c

With these patches the web replay tests are running and (usually) green on treeherder.

Comment 8

[Brian Hackett [Laid off!]]

Attachment: Part 7 - Avoid prematurely marking all threads as idle when there are in flight call events.

Fix a crash that happened while recording due to the main thread treating all threads as idle even though there were in flight off thread call events (these call events are a kind of tricky mechanism that allow the main thread to record snapshots even if other threads are doing blocking system calls like kevent).

https://hg.mozilla.org/projects/ash/rev/99de8008418f370c5be411c9bc113885f00facc2

With this change the web replay mochitests are green about 90% of the time.

Comment 9

[Brian Hackett [Laid off!]]

Closing this bug, all the changes here will be reviewed in separate bugs dependent on bug 1422587.