Re: Checkpoint taking long

On Tue, 20 Jul 2021 at 12:47, Yambu <hyambu(at)gmail(dot)com> wrote:

> Hello
>
> Can i kill a checkpoint process that has been running for 8hrs, below are
> from pg_stat_activity
>
>
> wait_event_type=IO
> wait_event=DataFileFlush
> backend_type=checkpointer
>
>
Is/was the backend storage having any issues, fsync slow?
i think killing checkpointer should be ok, as it will run again when it
reaches the thresholds of checkpoint_timeout or max_wal_size.
DataFileFlush: Waiting for a relation data file to reach durable storage.

as the checkpoint is basically doing the below, it should not be a problem
to kill it if is long running.
also , do you see this multiple times? if yes, can you try enabling
log_checkpoints so it would give an idea of how many buffers were written,
so as to help understand why it was waiting forever on IO.

1. Identify all the dirty pages in shared buffers.
2. Write the dirty pages to the respective files.
3. Issue fsync() on all the modified files to disk.

--
Thanks,
Vijay
Mumbai, India