Weird push problem

[Scott Palmer]

On 2011-08-21, at 11:16 PM, Tony Mechelynck wrote:

> On 14/08/11 03:49, Scott Palmer wrote:
>> 
>> On 2011-08-12, at 5:18 PM, Mads Kiilerich wrote:
>> 
>>> On 08/12/2011 10:54 PM, Scott Palmer wrote:
>>>> I'm trying to push changes to a repo I'm serving up with hgweb…  I haven't booted up the server I'm pushing to in a while, but it was working fine the last time I tried… the thing that would be different is that now I'm pushing from a more recent client version.
>>>> 
>>>> hg push
>>>> pushing to http://private_repo_URL_here/
>>>> searching for changes
>>>> 
>>>> .. does NOT ask for authorization - this is expected, it should be cache with mercurial_keyring.
>>>> .. now "sending" progress bar counts out to 12526
>>>> .. after that phase  is complete it suddenly asks for a password:
>>> 
>>> ...
>>> 
>>>> $ hg --version
>>>> Mercurial Distributed SCM (version 1.9.1+2-9991f8b19ff3)
>>>> (see http://mercurial.selenic.com for more information)
>>> 
>>> Patrick has made some important fixes in this area since 1.9.1 - see http://mercurial.selenic.com/bts/issue2885 . Please try again with the tip of the stable branch. It will probably either fix your issues or at least make them look different ;-)
>>> 
>>> He also made some fixes for keyring - make sure you also have these.
>>> 
>>> /Mads
>> 
>> Okay, this is getting interesting…
>> So I disabled the keyring extension and tried with:
>> Mercurial Distributed SCM (version 1.9.1+9-2ef2d3a5cd2d)
>> 
>> The problem was the same (with the exception of asking for credentials initially)
>> 
>> Now I am back home and not using that high-latency satellite network connection and I tried again with the same version of Mercurial, still with the keyring extension disabled, from the same repo on the same laptop - it worked perfectly.
>> 
>> I still don't get it.
>> 
>> [here's the last failure]
>> Scott-Palmers-MacBook-Pro:AlphaNorthAsset.com scott$ hg push
>> pushing to http://[private url]/
>> http authorization required
>> realm: analogideas
>> user: scott.palmer
>> password:
>> searching for changes
>> sending [<=>                               ] 43841/12526 -1972sabort: authorization failed
>> 
>> [I get home and try again - in same console window I re-run the last command]
>> 
>> Scott-Palmers-MacBook-Pro:AlphaNorthAsset.com scott$ hg push
>> pushing to http://[private url]/
>> http authorization required
>> realm: analogideas
>> user: scott.palmer
>> password:
>> searching for changes
>> remote: adding changesets
>> remote: adding manifests
>> remote: adding file changes
>> remote: added 10 changesets with 45 changes to 25 files
>> Scott-Palmers-MacBook-Pro:AlphaNorthAsset.com scott$ hg --version
>> Mercurial Distributed SCM (version 1.9.1+9-2ef2d3a5cd2d)
>> 
>> Yet the web front-end seemed to be working fine when I was on the other connection.  I got a log of the remote repo without a problem.
>> 
>> Scott
> 
> If it succeeds on your usual connection and fails on a slow satellite connection, with no other changes, then perhaps the failure is due to a timeout (or similar) error because the connection was too slow for the server and/or for the client?
> 

The failure was far too deterministic.  The other internet connection wasn't really slow, 1Mbit, just high latency, 1 second ping.  I checked server logs when I got back and there was nothing useful that I found in them. (No errors)

I would have also expected a reasonable error message from Mercurial if there was a simple timeout.

Cheers,

Scott