Closed Bug 887753 Opened 6 years ago Closed 6 years ago

Server not found after reconnecting to etherpad


(Core :: Networking, defect, major)

Windows 7
Not set





(Reporter: mayhemer, Assigned: mcmanus)




(1 file, 4 obsolete files)

I always just sleep my machine, never shutdown.  After wake up (after some 12 hours, or even less!) when I go to the tab with etherpad, there is a popup telling me to Reconnect.  I press the Reconnect button and get regular error page "Server not found".

Wireshark tells me we do a request for '' and get answer:

12	13:51:09.702044	fe80::219d:2101:836a:1eec	fe80::1	DNS	100	Standard query 0x220b  A

16	13:51:10.242770	fe80::1	fe80::219d:2101:836a:1eec	DNS	218	Standard query response 0x220b  A

18	13:51:10.700628	DNS	80	
Standard query 0x2594  A

19	13:51:10.711489	DNS	198	Standard query response 0x2594  A

I don't have a NSPR log, but I can create one if necessary.

100% reproducable for me.
Severity: normal → major
honza, jst managed to get a nspr log for something very similar sounding during the workweek. Its got a bunch of cookies/etc in it so I won't attach it here but jst will probably let me share it if you need it.

however, we might not need to because I've at least got a guess at what is going on from the log. I can't repro it - but I have seen it once or twice.

If you can repro it can you try out a try build with the potential fix?

(try is closed right now, I'll update the bug later with a cset)
Duplicate of this bug: 888082
tagging NI as I need someone from {jst, dougt, mayhemer} to evaluate try build and see if it fixes this issue.
Flags: needinfo?
So far inconclusive.  I always hit this when I sit to my machine new day and wake it and go to the tab with etherpad.  I left your build running during the night and tried exactly those steps.  It worked.  But then I tried to do the same in my regular profile with Aurora and it worked too.  So it's hard to say if it's fixed or not.

I'll do more attempts.
Flags: needinfo?
Patrick, I can no longer reproduce on my regular profile with Aurora :(  Bitch...

So I cannot confirm the fix.
I've ran a stock nightly build side by side with Patrick's try build today and I've seen the error show up 3 times so far in the stock nightly build, and zero times in the try build!
post patches - inquiring minds want to know!
Attachment #778760 - Attachment is obsolete: true
Attachment #778763 - Flags: review?(sworkman)
Attachment #778761 - Attachment is obsolete: true
Attachment #778762 - Attachment is obsolete: true
Attached patch 887753Splinter Review
jst's nspr log suggests a very quick dns failure. I suspect that's the inappropriate use of a negative cache entry. Its at least consistent.

negative entries are important to prefetch - because they can see a LOT of invalid names quickly and really jam up the q's with redundant references to them. (we consilidate overlapping requests, but it is still a problem without the negative cache)

So this patch just says to only use negative cache entries for < HIGH (i.e. speeculative) priority requests.. as a matter of hygiene it also dooms negative entries if they are used for a connection attempt.
Attachment #778764 - Flags: review?(sworkman)
Attachment #778763 - Attachment is obsolete: true
Attachment #778763 - Flags: review?(sworkman)
heh.. bzexport gave me an error each time I ran the command but it attached the diff anyhow each time (which I didn't notice). Sorry for the bugspam.
Comment on attachment 778764 [details] [diff] [review]

Review of attachment 778764 [details] [diff] [review]:

Had a good chat with Pat over email about this. Patch looks good to me r=me.
Attachment #778764 - Flags: review?(sworkman) → review+
Assignee: nobody → mcmanus
Closed: 6 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla25
Duplicate of this bug: 898564
Depends on: 923458
You need to log in before you can comment on or make changes to this bug.