Issues with Axigen Connector & HAProxy

Is anyone using HAProxy in front of their servers and using the Axigen Outlook Connector?

We find that when we go through the HA, the Outlook Connector continuously disconnects and reconnects. It will eventually finish the sync, but for large accounts it can be quite slow, and is very frustrating.

The logs are full of various errors, though some might just be irrelevant. The one error I have confirmed only appears when using the HA is:

BG [11:35:06.526]{12500, 16808 - folderWatcher }: Notifications received
WARN[11:35:06.526]{12500, 16808 - folderWatcher }: Error reading notification data: 3
WARN[11:35:06.526]{12500, 16808 - folderWatcher }: Abort notifications connection
WARN[11:35:06.526]{12500, 16808 - folderWatcher }: Retry opening notifications connection in 5 seconds
INFO[11:35:06.554]{12500, 10136 - connChecker }: Remote connection error state; recheck in one second
WARN[11:35:06.771]{12500, 14188 - syncProgress }: GetSyncStatus(000001B1BB5946C0, syncStatus, syncStatusReader, noWait=yes): signal error for job 4
INFO[11:35:07.555]{12500, 10136 - connChecker }: Remote connection error state; recheck in one second> WARN[11:35:07.766]{12500, 14188 - syncProgress }: GetSyncStatus(000001B1BB5946C0, syncStatus, syncStatusReader, noWait=yes): signal error for job 4

If I bypass the HA those errors all go away, but we sort of need the load balancer to work, so hoping someone else has experienced this!

HIya,

have you checked your time out settings on HAProxy?
If an IMAP Session times out too quick, OLK will have to reconnect and start all over again.

Google “haproxy imap time out” :slight_smile:

Cheers
JK

Hello Brad,

I second Jeroen hint.

Could you check if your HAProxy setup is able to keep opened IMAP sessions for more than 15-20 minutes without closing them due to inactivity?

This could be tested like:

$ time openssl s_client -starttls imap -crlf -connect your.email.server:143
1 login test.user@test.domain test.password
1 OK LOGIN completed
2 idle
+ Expecting DONE

We have seen a GCP implementation where when using the default LB setup the inactive sessions where “cut” after maximum 30 seconds of inactivity - and the result was breaking the usage of IMAP “idle” commands and ActiveSync push synchronizations.

HTH,
Ioan

2 Likes