Comments on: A Dive Into MySQL Multi-Threaded Replication

By: padam rakesh

padam rakesh — Tue, 13 Dec 2022 20:35:51 +0000

Thanks for blog..Even we use MTR also we see replication lag and app is doing 1000 rows/sec

[ 13-12-22 07:24:34 xxx ] > mysqlbinlog mysql-relay-bin.010530 | grep last_committed | awk ‘{ print $11″ “$12}’ | more
last_committed=0 sequence_number=1
last_committed=1 sequence_number=2
last_committed=1 sequence_number=3
last_committed=3 sequence_number=4
last_committed=4 sequence_number=5
last_committed=5 sequence_number=6
last_committed=6 sequence_number=7
last_committed=7 sequence_number=8
last_committed=8 sequence_number=9
last_committed=9 sequence_number=10
last_committed=10 sequence_number=11

From Master Binary log:

mysqlbinlog mysql-bin.003487 | grep last_committed | awk ‘{ print $11” “$12}’ | more
last_committed=0 sequence_number=1
last_committed=1 sequence_number=2
last_committed=2 sequence_number=3
last_committed=3 sequence_number=4
last_committed=4 sequence_number=5
last_committed=9461 sequence_number=9462
last_committed=9462 sequence_number=9463
last_committed=9463 sequence_number=9464
last_committed=9464 sequence_number=9465
last_committed=9465 sequence_number=9466
last_committed=9466 sequence_number=9467
last_committed=9467 sequence_number=9468

By: Yves Trudeau

Yves Trudeau — Wed, 06 Jul 2022 13:03:43 +0000

@Perica Yes, adding a sync wait time will slowdown the whole PXC cluster. With one node using a sync wait value, the majority ruling of Group replication would likely yield a better performance than the unanimous ruling of PXC.

By: Yves Trudeau

Yves Trudeau — Wed, 06 Jul 2022 13:00:01 +0000

@sjmudd for sure the durability aspect of xtradb cluster and group replication is quite different and interesting. I agree it would make an interesting post, maybe in the future.

By: Perica

Perica — Tue, 28 Jun 2022 08:28:58 +0000

I was wondering if these settings are safe to tune on 1 PXC node from a 3-node PXC cluster?
Would this effect in total cluster slowdown or with improving the lag on the replica?
Let’s say we have a standard 3 node PXC + 1 PS Replica connected to one of the PXC nodes.

By: sjmudd

sjmudd — Sat, 26 Mar 2022 08:59:03 +0000

Nice post.
A related but similar topic might be to discuss commit latency on something like xtradb cluster or group replication when spanning an AZ (to provide true HA) and how behavior may change with different load profiles. Replication is being pushed towards these technologies and this significantly changes a client’s perception of commit latency.
That would make for an interesting post