pt-table-checksum & innodb_stats_persistent_sample_page & Skipping table <tablename> because on the master it would be checksummed in one chunk but on these replicas it has too many rows

[Igor Sverkos]

Hi,

today I saw the following error for the first time while executing
"pt-table-checksum" v2.2.20:

> Skipping table <dbname>.<tablename> because on the master it would be checksummed in one chunk but on these replicas it has too many rows:
> 470585 rows on backup-db1
> The current chunk size limit is 463976 rows (chunk size=231988 * chunk size limit=2.0).

I read some bug reports and worked around passing '--chunk-size-limit
3' to the command.

However, I was wondering what's causing the difference so I looked
into the details:

> SELECT COUNT(*) AS n FROM <dbname>.<tablename>
> +--------+
> | n |
> +--------+
> | 470585 |
> +--------+


> SELECT TABLE_ROWS FROM information_schema.TABLES where table_name = '<tablename>';
> +------------+
> | TABLE_ROWS |
> +------------+
> | 426292 |
> +------------+

...so I learned about "innodb_stats_persistent_sample_page" which is set to 20.

Is it really the best way to just increase "chunk-size-limit"? Does
that work for larger tables as well? Isn't it a bad idea from
pt-table-checksum to rely on sampled values? Do you recommend a higher
"innodb_stats_persistent_sample_page" value when using
pt-table-checksum?


--
Regards,
Igor

[Lorraine Pocklington]

Hi 

I realise this doesn't address your last few questions, but there's a post about a similar error on the Percona forum with a couple of things you could check if you didn't see it already:

https://www.percona.com/forums/questions-discussions/percona-toolkit/24167-skipping-table-it-would-be-checksummed-in-1-chunk-but-replicas-have-too-many-rows

and it looks like others have found this post useful:  https://www.percona.com/blog/2015/08/12/mysql-replication-primer-with-pt-table-checksum-and-pt-table-sync/