pt-table-checksum Waits Indefinitely Checking for Max Chunk on Async PXC Node When Replication is not Properly Working

Reported by Jervin R on 2013-04-24
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Percona Toolkit
Undecided
Unassigned

Bug Description

Say, I have an async master with PC nodes as slave, if I have log_slave_updates turned off on the PXC nodes, replication stream does not work properly - this is a user case error, however for due diligence it should warn or bail for PXC cases like when log_slave_updates is off or for some reason that waiting for max chunk on the PXC nodes is not possible.

Im not sure what else to look for, but here's what I see on debug mode:

# pt_table_checksum:9551 8568 Checking slave diffs# pt_table_checksum:10533 8568 SELECT MAX(chunk) FROM `percona`.`checksums` WHERE db='test' AND tbl='audit' AND master_crc IS NOT NULL# pt_table_checksum:10550 8568 upxdbc01 max chunk: 1# pt_table_checksum:10550 8568 upxdbc02 max chunk: undef
# pt_table_checksum:10550 8568 upxdbc03 max chunk: undef
# pt_table_checksum:10583 8568 Sleep 0.25 waiting for chunks
# pt_table_checksum:10550 8568 upxdbc01 max chunk: 1
# pt_table_checksum:10550 8568 upxdbc02 max chunk: undef
# pt_table_checksum:10550 8568 upxdbc03 max chunk: undef
# pt_table_checksum:10583 8568 Sleep 0.5 waiting for chunks

This would go on and on.

[revin@forge msb_5_5_300]$ pt-table-checksum --version
pt-table-checksum 2.1.9

Daniel Nichter (daniel-nichter) wrote :

Is it just a matter of checking that log_slave_updates is off? Is this a PXC bug because it seems like log_slave_updates=OFF wouldn't affect regular replication or ptc?

tags: added: pt-table-checksum pxc
Changed in percona-toolkit:
status: New → Confirmed
Jervin R (revin) wrote :

Daniel,

If a PXC cluster is replicating from an async master, log_slave_updates is required on the node acting as async slave. If this is turned off, events from async master is only replicated to that node not the whole cluster - as such ptc have not knowledge of this and waits indefinitely.

I think this can be fixed with ptc knowing the topology and this requirement - the requirement for log_slave_updates I think is never gonna change.

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers