Re: Inconsistency in reporting checkpointer stats

On 2024/10/01 3:33, Fujii Masao wrote:
>
>
> On 2024/09/22 20:44, Nitin Jadhav wrote:
>>> +                                               CheckpointStats.ckpt_slru_written,
>>> +                                               (double) CheckpointStats.ckpt_slru_written * 100 / NBuffers,
>>>
>>> I don't think NBuffers represents the maximum number of SLRU buffers.
>>> We might need to calculate this based on specific GUC settings,
>>> like transaction_buffers.
>>
>> Great observation. Since the SLRU buffers written during a checkpoint
>> can include transaction_buffers, commit_timestamp_buffers,
>> subtransaction_buffers, multixact_member_buffers,
>> multixact_offset_buffers, and serializable_buffers, the total count of
>> SLRU buffers should be the sum of all these types. We might need to
>> introduce a global variable, such as total_slru_count, in the
>> globals.c file to hold this sum. The num_slots variable in the
>> SlruSharedData structure needs to be accessed from all types of SLRU
>> and stored in total_slru_count. This can then be used during logging
>> to calculate the percentage of SLRU buffers written. However, I’m
>> unsure if this effort is justified. While it makes sense for normal
>> buffers to display the percentage, the additional code required might
>> not provide significant value to users. Therefore, I have removed this
>> in the attached v6 patch.
>
> +1
>
> Thanks for updating the patch! It looks good to me.
> Barring any objections, I will commit this patch.

Before committing the patch, I revised it with the following changes:

- I added "shared" to the description of pg_stat_checkpointer.buffers_written
to clarify that it tracks shared buffers.
- In the checkpoint log message, I replaced "slru" with "SLRU" for consistency,
as uppercase is typically used for SLRU.
- I updated the commit message.

Regards,

--
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION