> Just to check I have this straight:
> - Proxmox cluster using GlusterFS for storage
> - bricks on Proxmox nodes
> - Linux VM's running on Proxmox Nodes
> - InnoDB running on the linux vms

Yes, that's exactly it.

> When one of the proxmox nodes crashes (Power outage?) the InnoDB 
> database is hosed?

Not always, but often. Most of the time enabling force recovery fixes it,
but not always, I had to export, wipe and then re-import everything a few times
to fix innodb.

> We run multiple MS SQL servers in the same setup and a few mysql, never 
> had that problem with them after server outages.
> - Replica 3?
> - Could you post your gluster info?
> - Whats the underlying filesystem for the bricks? ZFS? What sync mode 
> does it have set?
> - Whats the KVM cache mode?

The KVM cache is directsync, I was using none before and I started putting
directsync hoping to fix that problem.
The bricks are on XFS, but I also tried ext4. I didn't try ZFS because I have
no idea how that even works.

Here is the config of one of the clusters :

Volume Name: VMs
Type: Replicate
Volume ID: c5272382-d0c8-4aa4-aced-dd25a064e45c
Status: Started
Number of Bricks: 1 x 3 = 3
Transport-type: tcp
Brick1: ips4adm:/mnt/storage/VMs
Brick2: ips5adm:/mnt/storage/VMs
Brick3: ips6adm:/mnt/storage/VMs
Options Reconfigured:
performance.readdir-ahead: on
cluster.quorum-type: auto
cluster.server-quorum-type: server
network.remote-dio: enable
cluster.eager-lock: enable
performance.quick-read: off
performance.read-ahead: off
performance.io-cache: off
performance.stat-prefetch: off
features.shard: on
features.shard-block-size: 64MB
cluster.data-self-heal-algorithm: full
network.ping-timeout: 15

The data-self-heal-algorithm was advised by Krutika I believe,
when we were having huge heal problems.
Appart from that, everything works fine. fsck when the VM starts after
a crash usually finds a few things, but nothing big, everything boots
and nothing is missing. Only MySQL, from time to time.

I imagine you might run into the problem someday too, we're just having
a lot more crashes than the average user for reasons I mentionned in
another mail.

Kevin Lemonnier
PGP Fingerprint : 89A5 2283 04A0 E6E9 0111

Attachment: signature.asc
Description: Digital signature

Gluster-users mailing list

Reply via email to