[Gluster-users] Quick and small file read/write optimization

[Pedro Costa]

Hi,

I've a  1 x 3 replicated glusterfs 4.1.5 volume, that mounts using fuse on each server into /www for various Node apps that are proxied with nginx. Servers are then load balanced to split traffic. Here's the gvol1 configuration at the moment:

Volume Name: gvol1
Type: Replicate
Volume ID: 384acec2-XXXX-40da-YYYY-5c53d12b3ae2
Status: Started
Snapshot Count: 0
Number of Bricks: 1 x 3 = 3
Transport-type: tcp
Bricks:
Brick1: vm0:/srv/brick1/gvol1
Brick2: vm1:/srv/brick1/gvol1
Brick3: vm2:/srv/brick1/gvol1
Options Reconfigured:
cluster.strict-readdir: on
client.event-threads: 4
cluster.lookup-optimize: on
network.inode-lru-limit: 90000
performance.md-cache-timeout: 600
performance.cache-invalidation: on
performance.cache-samba-metadata: on
performance.stat-prefetch: on
features.cache-invalidation-timeout: 600
features.cache-invalidation: on
transport.address-family: inet
nfs.disable: on
performance.client-io-threads: on
storage.fips-mode-rchecksum: on
features.utime: on
storage.ctime: on
server.event-threads: 4
performance.cache-size: 500MB
performance.read-ahead: on
cluster.readdir-optimize: on
cluster.shd-max-threads: 6
performance.strict-o-direct: on
server.outstanding-rpc-limit: 128
performance.enable-least-priority: off
cluster.nufa: on
performance.nl-cache: on
performance.nl-cache-timeout: 60
performance.cache-refresh-timeout: 10
performance.rda-cache-limit: 128MB
performance.readdir-ahead: on
performance.parallel-readdir: on
disperse.eager-lock: off
network.ping-timeout: 5
cluster.background-self-heal-count: 20
cluster.self-heal-window-size: 2
cluster.self-heal-readdir-size: 2KB

On each restart the apps delete a particular folder and rebuild it from internal packages. On one such operation on a particular client to the volume I get repeated logs hundreds of times for the same guid even, sometimes:

[2018-10-09 13:40:40.579161] W [MSGID: 114061] [client-common.c:2658:client_pre_flush_v2] 0-gvol1-client-2:  (7955fd7a-3147-48b3-bf6a-5306ac97e10d) remote_fd is -1. EBADFD [File descriptor in bad state]
[2018-10-09 13:40:40.579313] W [MSGID: 114061] [client-common.c:2658:client_pre_flush_v2] 0-gvol1-client-2:  (0ac67ee4-a31e-4989-ba1e-e4f513c1f757) remote_fd is -1. EBADFD [File descriptor in bad state]
[2018-10-09 13:40:40.579707] W [MSGID: 114061] [client-common.c:2658:client_pre_flush_v2] 0-gvol1-client-2:  (7ea6106d-29f4-4a19-8eb6-6515ffefb9d3) remote_fd is -1. EBADFD [File descriptor in bad state]
[2018-10-09 13:40:40.579911] W [MSGID: 114061] [client-common.c:2658:client_pre_flush_v2] 0-gvol1-client-2:  (7ea6106d-29f4-4a19-8eb6-6515ffefb9d3) remote_fd is -1. EBADFD [File descriptor in bad state]

I assume this is probably because the client didn't catch up with the previous delete? I do control the server (client to the gluster volume) that the restart occurs, and I prevent having more than one rebuilding the same app at the same time, which makes these logs odd.

I've implemented the volume options above after reading most of the entries in the archive here over the last few weeks, but I'm not sure what else to tweak because other than the restart of the apps it is working pretty well.

If there's any input you may have to help on this particular scenario I'd be much appreciated.

Thanks,
P.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20181009/3b8a18fa/attachment.html>