[HTTPCLIENT-2081] Intermittent Connection reset errors while talking to IBM COS

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Not A Problem
• Affects Version/s: 4.5.12
• Component/s: HttpClient (classic)
• Assignee: Unassigned
• Reporter: Philipp Nanz
• Created: 25/May/20 06:32
• Updated: 31/May/20 14:25
• Resolved: 25/May/20 18:03

Description

Hi there,

we are currently using HttpClient for downloading blobs off IBM COS, which is basically a S3-like cloud storage.

Sometimes we are seeing that HttpClient is giving up with "connection reset" errors after doing its retries.

A quick Jira search showed that there were similar issues reported recently, but not investigated further (HTTPCLIENT-1954, HTTPCLIENT-2049).

I have attached a log file including the full wire protocol.

There is a couple of things that seems fishy: first of all, there is a read timeout being reported after 3 ms or so. On the subsequent retries, I suspect that the same connection is being reused and hence the peer sends the RST packets. But that is just my wild speculation, because I cannot read much from these logs.

Since this is happening only ocassionally, I'm guessing at a timing related issue. Hence, my next step is to tweak the retry handler in a way that it waits for a small amount of time. I will report back my findings.

If there is anything else you need, please let me know. I will be happy to provide anything that is necessary to further invbestigate this issue.