Closed Bug 1236376 Opened 8 years ago Closed 8 years ago

[Gij] RangeError: port should be >= 0 and < 65536: 65537, RangeError: port should be >= 0 and < 65536: 65536 ... Not connected. To write data you must call connect first

Categories

(Testing Graveyard :: JSMarionette, defect)

Product:
Testing Graveyard
Component:
JSMarionette
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

(feature-b2g:2.6+)

Status:
RESOLVED FIXED
Milestone:
2.6 S5 - 1/15
Project Flags:
feature-b2g 2.6+

People

(Reporter: mikehenrty, Assigned: mikehenrty)

References

Details

(Keywords: intermittent-failure, Whiteboard: [systemsfe][MJS] [CI])

Attachments

(2 files)

[gaia] mikehenrty:bug-1236376-use-more-ports > mozilla-b2g:master
46 bytes, text/x-github-pull-request
gmarty
: review+
Details | Review
[gaia] mikehenrty:bug-1236376-use-more-reland > mozilla-b2g:master
46 bytes, text/x-github-pull-request
gmarty
: review+
Details | Review 
This problem seems to happen for all Marionette JS tests sporadically. For example [1]. The problem seems to be that emptyPort can't find a valid open port. We should examine what is happening here.

1.) https://treeherder.mozilla.org/logviewer.html#?job_id=19192954&repo=mozilla-inbound
Assignee: nobody → mhenretty
Keywords: intermittent-failure
Whiteboard: [systemsfe][MJS]
Hah, so by default empty-port would use a MAX_PORT of 65537 (65535 being the true valid max), so 2/5507 times it would pick a completely illegal port.  By specifying the max port now you help eliminate that foolishness.

Meta-wise, the randomly try a port solution is strictly worse than just binding to a 0 port to see what the OS allocates since the empty-port implementation could waste a lot of time randomly picking ports that are already taken.

Better yet would be to change the port-assigning dance by having the thing that ends up listening do the bind and then provide it over a named pipe or writing to a file or stdout or something.  That would avoid port allocation races, etc.
Summary: [Gij] RangeError: port should be >= 0 and < 65536: 65537 ... Not connected. To write data you must call connect first → [Gij] RangeError: port should be >= 0 and < 65536: 65537, RangeError: port should be >= 0 and < 65536: 65536 ... Not connected. To write data you must call connect first
Comment on attachment 8703459 [details] [review]
[gaia] mikehenrty:bug-1236376-use-more-ports > mozilla-b2g:master

I received verbal ok from Aus and Gareth yesterday about this bug. Sending over to Guillaume to do final review.
Attachment #8703459 - Flags: review?(gmarty)
Comment on attachment 8703459 [details] [review]
[gaia] mikehenrty:bug-1236376-use-more-ports > mozilla-b2g:master

Looks good to me!
Attachment #8703459 - Flags: review?(gmarty) → review+
master: https://github.com/mozilla-b2g/gaia/commit/16f4490e8208cc69b8e522fb468e9617eee85d53
Status: NEW → RESOLVED
Closed: 8 years ago
Resolution: --- → FIXED
(In reply to Andrew Sutherland [:asuth] from comment #3)
> Hah, so by default empty-port would use a MAX_PORT of 65537 (65535 being the
> true valid max), so 2/5507 times it would pick a completely illegal port. 
> By specifying the max port now you help eliminate that foolishness.
> 
> Meta-wise, the randomly try a port solution is strictly worse than just
> binding to a 0 port to see what the OS allocates since the empty-port
> implementation could waste a lot of time randomly picking ports that are
> already taken.
> 
> Better yet would be to change the port-assigning dance by having the thing
> that ends up listening do the bind and then provide it over a named pipe or
> writing to a file or stdout or something.  That would avoid port allocation
> races, etc.

Thanks for this Andrew. I filed bug 1236855 about this.
This is causing lots of the following error:
[taskcluster:error] Task has been aborted prematurely. Reason: worker-shutdown
https://treeherder.mozilla.org/logviewer.html#?job_id=732216&repo=gaia-master

Hard to tell what's happening, but it's almost certainly caused by my patch here, so I'm going to back this out for now.
Reverted in master: https://github.com/mozilla-b2g/gaia/commit/7c74f6a655c44af47be9e328e65c47a9740ac19c
Status: RESOLVED → REOPENED
Resolution: FIXED → ---
feature-b2g: --- → 2.6+
Comment on attachment 8704167 [details] [review]
[gaia] mikehenrty:bug-1236376-use-more-reland > mozilla-b2g:master

Let's just use querySelector instead in both places to be safe. Guillaume, wanna take a look?
Attachment #8704167 - Flags: review?(gmarty)
Comment on attachment 8704167 [details] [review]
[gaia] mikehenrty:bug-1236376-use-more-reland > mozilla-b2g:master

Looks perfect to me!
Attachment #8704167 - Flags: review?(gmarty) → review+
(In reply to Michael Henretty [:mhenretty] from comment #27)
> Ok, this test should no longer be a problem here.

Sorry, that comment was meant for bug 1236329. I am not as confident of this being the final fix, but it should at least move us in the right direction.
Whiteboard: [systemsfe][MJS] → [systemsfe][MJS] [CI]
Target Milestone: --- → 2.6 S5 - 1/15
Look at the brasstacks link in comment 30, this hasn't happened since Jan 7, which would be enough time for my changes to propagate to all trees. Marking as fixed.
Status: REOPENED → RESOLVED
Closed: 8 years ago → 8 years ago
Resolution: --- → FIXED
Product: Testing → Testing Graveyard
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: