Re: [PATCH] jbd2: Increase IO priority of checkpoint.

[Julian Sun <sunjunchao@bytedance.com>]

在 2025/8/27 19:15, Jan Kara 写道:
> On Wed 27-08-25 12:57:32, Julian Sun wrote:
>> 在 2025/8/27 04:55, Ritesh Harjani (IBM) 写道:
>>> Julian Sun <sunjunchao@bytedance.com> writes:
>>>
>>>> In commit 6a3afb6ac6df ("jbd2: increase the journal IO's priority"),
>>>> the priority of IOs initiated by jbd2 has been raised, exempting them
>>>> from WBT throttling.
>>>> Checkpoint is also a crucial operation of jbd2. While no serious issues
>>>> have been observed so far, it should still be reasonable to exempt
>>>> checkpoint from WBT throttling.
>>>>
>>>
>>> Interesting.. I was wondering whether we were able to observe any
>>> throttling for jbd2 log writes or for jbd2 checkpoint?
>>> Maybe It would have been nice, if we had some kind of data for this.
>>
>> Good idea. But AFAICS wbt lacks of such a obversation mechanism now..>
> 
> Well, I guess Ritesh meant some test case which is reproducing the
> situation where jbd2 log writes get stalled.

Ah, thanks for the clarification. It may be a little bit difficult to 
reproduce it manually, I’ll give it a shot but I’m not sure if I can do 
it..>
>>> BTW - does it make sense for fastcommit path too maybe for non-tail
>>> fc write requests? I think it uses ext4_fc_submit_bh().
>>
>> Yeah, I think so.
>> After a rough check of the code, the following code paths may result in high
>> latency or even task hangs:
>>    1. fastcommit io is throttled by wbt or other block layer qos policies.
>>    2. jbd2_fc_wait_bufs() might wait for a long time while
>> JBD2_FAST_COMMIT_ONGOING is set in journal->flags, and then
>> jbd2_journal_commit_transaction() waits for the JBD2_FAST_COMMIT_ONGOING bit
>> for a long time while holding the write lock of j_state_lock.
>>    3. start_this_handle() waits for read lock of j_state_lock which results
>> in high latency or task hang.
>>
>> Hi, Jan, please correct me if I'm missing anything.
> 
> I think you're correct. In fact ext4_fc_commit() does modify current
> process' io priority to match that of jbd2 thread so it would be logical
> to match IO submission flags as well.

Thanks, I will send a patch to improve here.>
> 								Honza


-- 
Julian Sun <sunjunchao@bytedance.com>