Ask Your Question

Revision history [back]

Please pardon all these questions, if you have already answered them (and so I just could not dig through the history to pull out the answers well enough).

Is it the same error as what you first reported? 10s ChunkReadTimeout? Can you post them?

Can I assume that each node in this swift configuration is running proxy, account, container, and object servers? I think the object server worker counts are at 8, is that still the case? Did you try to set max_clients lower, raising the number of workers? Did you try ThreadPools (and for that matter, what version of Swift are you running)?

Can you post the output of "xfs_info" on a disk? We have not seen much benefit, if at all, with inode sizes of 1,024. Additionally, for large disks, the default allocation groups that XFS uses can be kinda small (by default 4 allocation groups per 1 TB, so hoping it is at 16 or higher for your disks). And depending on how big your partitions on a disk grow to, you might benefit from a larger name size, 8192 or higher depending.

Are you using 1ge or 10ge networking? And do you a single nic for all traffic, front and backend, or do you separate your client traffic from your inter-swift-node traffic.

I don't think memcache is a factor here, but curious about where your memcache servers are located?

Finally, can you post a one minute capture of just your proxy logs, scrubbed?