Project

General

Profile

Fix #7389

ZeroRPC connections stuck open when running inside Apache

Added by John Spray about 10 years ago. Updated about 10 years ago.

Status:
Resolved
Priority:
Normal
Assignee:
Category:
Backend (REST API)
Target version:
% Done:

0%

Source:
other
Tags:
Backport:
Reviewed:
Affected Versions:
ceph-qa-suite:
Crash signature (v1):
Crash signature (v2):

Description

From an overnight test (damn, it sure it is a good idea to run these!)

Production mode instance falls over after running for some time, turns out Apache has 10s of thousands of TCP sockets open. They look to be the ZeroMQ connections to cthulhu.

Associated revisions

Revision 4b5a9e66 (diff)
Added by John Spray about 10 years ago

calamari_rest: close ZeroRPC connections

This partly fixes #7389. Even with this patch,
we still get zombie connections if HTTP requests
are aborted by the client (e.g. if you pump
the refresh button in the browser).

I don't actually understand why these connections
weren't getting closed on garbage collection to
begin with, suspect it's some interplay between
gevent and apache.

Revision 19948d16 (diff)
Added by John Spray about 10 years ago

calamari_web: Close RPC connection used in dashboard() view

Fixes: #7389

History

#2 Updated by John Spray about 10 years ago

  • Status changed from New to Resolved

Fixed at 19948d16cada0ab5d85c90b5813a1457e6577182

Also available in: Atom PDF