Map/Reduce frequently throws riak.RiakError: 'Socket returned short packet length 3 - expected 4' about riak-python-client

[evanmcc]

I am not able to replicate this. Do you have a good simplified case that can replicate it semi-reliably?

[yizha]

Well, it's a small project and it is closed (you see it has been two months) and the dev env is gone too so I cannot replicate it myself or provide a simple case (cannot get those data back). But I think it is apparent that it tries to get 4 bytes by calling conn.recv(...) only once which is not correct.

[evanmcc]

I guess the while is what worries me here, since there's no guarantee that it'll terminate in a timely fashion. I'll think about it.

[evanmcc]

closing this as I still haven't been able to replicate it.

[numan]

Can we reopen this and look at this again. I actually get this quite often in production. I'm not sure what the solution is, but maybe we can open it up again for discussion,

[numan]

Seems to happen mostly during mapreduce calls for us.

[hazen]

@numan - I'll ask the same thing Evan did last year: Do you have a small test case which would allow us reproduce this? Also which version of the Python client are you using?

[numan]

Unfortunately not. It occurs randomly once in a while, with no predictable pattern. I'm continuing to see if I can find a test case while allows me to reproduce this.

I'm using version 2.0.2

[lexual]

I too don't have a reproducible test case.

But we're seeing this in production randomly but consistently when running Map/Reduce jobs (when using protobuf).

See a csv of errors when we've seen this here:

http://pastebin.com/5UMgQkB1

[lexual]

Running 2.0.2 version of python bindings.

Running on Ubuntu in AWS.

I too would like this issue re-opened to investigate it further.

In another usecase, our solution has had to be to switch back to HTTP so we don't experience this error.

cc @mafrosis

[seancribbs]

Resolved in #361