Since https://github.com/systemd/systemd/commit/a327431bd168b2f327f3cd422379e213c643f2a5 released in system v236 Type=notify service can now advice to the systemd service manager they are still working to avoid the service timing out.
The use of EXTEND_TIMEOUT_USEC= on older services has no effect and is therefore compatible.
This needs to be included in (feel free to correct/extend):
- buffer pool dump - buf_dump (storage/innobase/buf/buf0dump.cc)
- redo log recovery - log_group_read_log_seg (storage/innobase/log/log0log.cc)
- undo recovery - recv_recover_page_func (storage/innobase/log/log0recv.cc)
- change buffer?
- merge buffer?
I was planning on making the 15 seconds of recv_sys_t->report() more general with respect to interval, and use a define INNODB_REPORT_INTERVAL (include/univ.i?) as the basis for this form of watchdog. I'd send notify messages of INNODB_REPORT_INTERVAL * 2 as an acceptable margin.
- galera SST scripts - donor and recipient
Any other server/engine slow points to account for?
Target 10.3 and then look at a backport?