It’s chain replication A->B->C->A and pt-table-checksum works great only with STATEMENT format for this particular setup AFAIK. You can read further here on LIMITATION section. [url]pt-table-checksum — Percona Toolkit Documentation
Checksumming relies on binlog_format=STATEMENT. B receives changes from it’s Master A, which will be STATEMENT (pt-table-checksum will force this into STATEMENT binlog_format in it’s session) B executes the checksumming queries. So far so good.
But then B will also write the changes to it’s own binary log, so it can replicate to C but It will write checksum queries as ROW events in binary log so It won’t replicate to C.
pt-table-checksum only works properly when STATEMENT based formatting is used (to generate the checksum), if it’s ROW/MIXED based, the row change to the checksums table is is just replicated from it’s master, no checksumming happens…