Oneshot push replication creating ~1500 documents crashes SG 2.6

Mobile Sync Gateway
1

SG 2.6 (Couchbase Sync Gateway/2.6.0(127;b4c828d) CE)
Ubuntu 16.04
CBL 2.6
Android

This issue might be related to CBL 2.6: Does SG need more RAM? Initially only? and a fix might be in the next release, see post #9 in the same thread.

Attached are the goroutines at the time of the crash. The first goroutine (…20_28_33) is when a user crashed the SG. Second goroutine was me by testing the oneshot push replication (i.e. signing in to an account the first time). Heap and goroutine are downloaded and stored automatically as soon as the SG crashes/hangs. Then it’s restarted.

This issue is reproducible on my production cluster! It worked on my dev cluster - a lot less powerful, single user only. Production cluster might have around 200 concurrent sync users. So not too many. There is lots of unallocated RAM and IDLE compute resources available. Prod and dev have the same software installed.

Is this issue related to my other post?

Thanks!

(By SG crash I mean that all connected Android clients show the replication status BUSY. No changes are synced until the SG service is restarted.)

goroutine-2019_12_12___20_28_33.zip (6.1 KB) goroutine-2019_12_12___21_44_28.zip (6.6 KB)

2

Going over the posts I believe it is the same issue. Initially I thought that an already signed in user pushes up a lot of documents. Maybe even documents which are rejected by the SG function and hence this issue keeps appearing. Now I think it is new users who push up their entire local dataset (which they created themselves) for the first time. I guess it’s best to wait for the new SG release in 2020.

3

There were no more issues since upgrading to SG version 2.7.

1 Like