Re: Prolonged truncation phase during vacuum on toast table with repeated interruptions by lock waiters and a proposed POC patch

Hi hackers,

Following up on the recent discussion about prolonged VACUUM truncation phases on busy systems, I took a stab at extending the logging to add a little bit more visibility into the truncation process.

With this patch, when VACUUM is run with VERBOSE, it will now log the following:

1. Individual suspension messages with attempt count and cumulative elapsed time.
2. A summary at the end of the truncation process showing the total number of suspensions and the total elapsed time.

Example output:
  INFO:  table "toast_target_table": suspending truncate due to conflicting lock request (attempt 1, elapsed: 0.121 s)
  INFO:  table "toast_target_table": truncated 372096 to 370624 pages
  INFO:  table "toast_target_table": suspending truncate due to conflicting lock request (attempt 2, elapsed: 6.108 s)
  ...
  INFO:  table "toast_target_table": truncation completed after 8 suspension(s), total elapsed: 58.090 s

Perhaps, this can give developers the information needed to identify problematic truncation behavior and decide whether to use VACUUM (TRUNCATE OFF) for specific tables.

Additionally, tracking these stats in a table would be useful as well. Curious, if something like this is desirable by the community?

Proposal patch is attached. Some notes on the patch:

- Unclear if there is a best way to write regression tests for this, since it can be flaky. 

- There seems to be translation support for the copy. At least for the initial patch, I have left that and happy to discuss more if the patch feels viable. 

Thank you
Shayon