the puller scheduler's handling of broken netstrings is broken

[Michael Hudson-Doyle]

Thanks in part to a Twisted issue (http://twistedmatrix.com/trac/ticket/2851), the puller's handling of malformed netstrings doesn't actually work, as seen in OOPS-1219SMPM3.

It's a bit hard to see what to do about this, apart from fixing Twisted.

One possible cause of this issue are 'mirrorFailed' messages with extremely long (> 100k) explanations.

[Jonathan Lange (jml)]

OOPS-1219SMPM3 has the error specific to netstring handling.

OOPS-1219SMPM4 is presumably a later response to the same root error in the worker.

OOPS-1219SMPM01 lends support to Michael's theory of an extremely long mirrorFailed message (warning: OOPS report is very large) and OOPS-1219SMPM09 is from the same "branch".

[Tim Penhey (thumper)]

Michael, is this still a problem?

[Michael Hudson-Doyle (mwhudson)]

I think so, although it happens so extreeeeemely rarely we should lower the priority.

[Robert Collins (lifeless)]

FTR last seen - Last seen: Oct. 12, 2010, 2:19 p.m. OOPS-1746SMS382. This is a while ago now, but mwhudson did say very rare. @mwhudson, do you think its > 1 year frequency and still existing, or perhaps fixed by some twisted upgrade?

[Robert Collins (lifeless)]

Oh, and if jelmer completes his work to make the puller go away, using regular code imports, we can close this as no longer relevant to LP.

[Michael Hudson-Doyle (mwhudson)]

I'm pretty sure the bug is still there. Twisted has changed a bit so we could probably fix this by overriding a private method -- but we don't (and it's not clear that this would be a good idea -- changing twisted to make the method public would be better).

Agree that Jelmer is close to making this bug irrelevant.