ConnectionTimeout in high concurency

[David Hadas]

In a setup that includes a proxy and a separate backend storage server,
running swift-bench with:
    concurrency = 100
     object_size = 128000

produces failures. Log file suggests ConnectionTimeout is raised.
Reducing the concurrency to 5 on my system removes the problem.

[David Hadas (david-hadas)]

See:
https://github.com/eventlet/eventlet/issues/50

[David Hadas (david-hadas)]

Since Swift uses only poll and select we can create a modified derived class for both the poll and select hubs and solve the eventlet issue (while trying to push this to eventlet as well)

[Hisashi Osanai (osanai-hisashi)]

Hi David,

When I put an object that size is 256MB with 100 samples and 1
concurrency using swift-bench, I got following messages.

Jul 8 18:35:22 xxxxx proxy-server: ERROR with Object server 192.168.xx.87:6000/device0 re: Trying to get final status of PUT to /v1/AUTH_e76c621426b748b1acd32c2af359b59a/409bedd98f204d4e95854c0660350207_16/d5929b3fb1284f5f9e7b2ab75f2d0bac: Timeout (10s)

I found this bug report and checked whether there is the fix for
"Eventlet prefers timeout handling over other non blocking work #50"
in eventlet/hubs/poll.py (python-eventlet-0.14.0-1.el6.noarch).
The result is there is no fix for this.

The fix is:
https://github.com/eventlet/eventlet/commit/87845b039296bc405de88b7e9b2942597dc7ba7f

If possible, I would like to have your thought whether the output of above
timeout messages are same as your finding.