Retry to send request after receiving connection reset by peer.

[ppluciennik]

Code tries twice to send request and fails after that.
Problem is observed while running ansible playbooks. Resending request helps.

[silencev]

how about giving a argument to specify try count?

[coveralls]

Coverage decreased (-1.2%) to 66.667% when pulling 90d3a40 on ppluciennik:ppluciennik/crbp into b1a54bc on diyan:master.

[coveralls]

Coverage decreased (-1.2%) to 66.667% when pulling ca673ea on ppluciennik:ppluciennik/crbp into b1a54bc on diyan:master.

[coveralls]

Coverage decreased (-1.2%) to 66.667% when pulling ca673ea on ppluciennik:ppluciennik/crbp into b1a54bc on diyan:master.

[coveralls]

Coverage decreased (-1.2%) to 66.667% when pulling 90d3a40 on ppluciennik:ppluciennik/crbp into b1a54bc on diyan:master.

[coveralls]

Coverage decreased (-1.2%) to 66.667% when pulling 90d3a40 on ppluciennik:ppluciennik/crbp into b1a54bc on diyan:master.

[coveralls]

Coverage decreased (-1.2%) to 66.667% when pulling 90d3a40 on ppluciennik:ppluciennik/crbp into b1a54bc on diyan:master.

[ppluciennik]

@diyan: Is it feasible to integrate it soon?

[nitzmahone]

Blind retries at the send_message level can be extremely dangerous; you don't know if/how far the request got the way this is written right now. I know @dagwieers was playing with a much more targeted version of this behavior at one point and only doing it when it's known to be absolutely safe, but not sure where it stands. I'm going to close this one as a "no thanks" for now- might want to check with Dag about his stuff.

[nitzmahone]

Oh, it was the next one in the queue: #174

[dagwieers]

So my implementation is in #174 and currently has hard-coded retry count and sleep time. I would need advice on how we want to make this configurable.

We have been using this in production now, as it fixes a few corner-cases described in Ansible issue ansible/ansible#25532

Also, there may be certain HTTP errors with additional WinRM information indicating a retry would work, I remember we discussed this at AnsibleFest SF. I don't know if we have some more information related to this ?

[nitzmahone]

Potentially we will, if you're still hitting those- I plugged in a hacked-up version of the SOAP fault preservation code, so we'll know everything there is to know about 500s in 0.3.0.

[dagwieers]

We isolated the various requests that can fail and can be retried, #174 now implements what is possible. Please try this for your existing use-cases.