Project

General

Profile

Bug #10447

rx to radosgw is more than tx (more than ten times)

Added by Mustafa Muhammad about 9 years ago. Updated about 4 years ago.

Status:
Closed
Priority:
Normal
Assignee:
-
Target version:
-
% Done:

0%

Source:
Community (user)
Tags:
rgw
Backport:
Regression:
No
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
Crash signature (v1):
Crash signature (v2):

Description

Hi, I am using radosgw (using civetweb, but happens with apache too), rx to the gateway is more than 10 times the tx to the clients.

History

#1 Updated by Yehuda Sadeh about 9 years ago

  • Status changed from New to Need More Info

More information about what you're actually seeing, providing some logs maybe will help here. I honestly have no idea what you actually mean.

#2 Updated by Mustafa Muhammad about 9 years ago

I have a production ceph cluster, we use multiple rados gateways with it, the problem is, for each radosgw with high connection count, it receives huge amount of data from the cluster (compared to the data it forwards to the customer), this is very visible in civetweb (but I reported another issue about that, issue 10606 "civetweb keeps receiving data after client canceled download")
I couldn't see anything strange in the logs, how can I add more verbosity to give you useful log?

#3 Updated by Yehuda Sadeh about 9 years ago

So basically you're seeing that with civetweb, an aborted download is not detected?

#4 Updated by Mustafa Muhammad about 9 years ago

Yehuda Sadeh wrote:

So basically you're seeing that with civetweb, an aborted download is not detected?

Not just that, that is the other issue I reported in another report, here, even with apache when I have a large number of connections, the difference increases (sometimes to 10x).

#5 Updated by Casey Bodley about 4 years ago

  • Status changed from Need More Info to Closed

#6 Updated by Mustafa Muhammad about 4 years ago

Hi, I don't understand why this was just closed, nowadays we are using RHCS but it still happen, is there a fix that was merged or something?

Also available in: Atom PDF