Re: [RFC] mm/vmscan.c: avoid possible long latency caused by too_many_isolated()

[Hillf Danton <hdanton@sina.com>]

On Mon, 10 May 2021 16:03:06  Xing Zhengjun wrote:
>On 4/30/2021 2:43 PM, Hillf Danton wrote:
>> On Fri, 30 Apr 2021 13:33:57 +0800 Xing Zhengjun wrote:
>>>
>>> I use my compaction test case to test it, 1/10 ratio can reproduce 100ms
>>> sleep.
>>>
>>>   60) @ 103942.6 us |      shrink_node();
>>>
>>>   60) @ 103795.8 us |      shrink_node();
>> 
>> Thanks for your test.
>> 
>> In bid to cut the number of 100ms sleepers further down, add another place
>> for them to nap by flushing lru cache before falling in sleep, instead of
>> mulling why 50ms or 10ms is more adequate.
>> 
>> Alternatively, and simpler IMHO, take a 5ms nap one time until !tmi.
>> 
>> --- y/mm/vmscan.c
>> +++ x/mm/vmscan.c
>> @@ -118,6 +118,9 @@ struct scan_control {
>>   	/* The file pages on the current node are dangerously low */
>>   	unsigned int file_is_tiny:1;
>>   
>> +	unsigned int file_tmi:1; /* too many isolated */
>> +	unsigned int anon_tmi:1;
>> +
>>   	/* Allocation order */
>>   	s8 order;
>>   
>> @@ -2092,6 +2095,22 @@ static int current_may_throttle(void)
>>   		bdi_write_congested(current->backing_dev_info);
>>   }
>>   
>> +static void set_sc_tmi(struct scan_control *sc, bool file, int tmi)
>> +{
>> +	if (file)
>> +		sc->file_tmi = tmi;
>> +	else
>> +		sc->anon_tmi = tmi;
>> +}
>> +
>> +static bool is_sc_tmi(struct scan_control *sc, bool file)
>> +{
>> +	if (file)
>> +		return sc->file_tmi != 0;
>> +	else
>> +		return sc->anon_tmi != 0;
>> +}
>> +
>>   /*
>>    * shrink_inactive_list() is a helper for shrink_node().  It returns the number
>>    * of reclaimed pages
>> @@ -2109,11 +2128,23 @@ shrink_inactive_list(unsigned long nr_to
>>   	enum vm_event_item item;
>>   	struct pglist_data *pgdat = lruvec_pgdat(lruvec);
>>   	bool stalled = false;
>> +	bool drained = false;
>>   
>>   	while (unlikely(too_many_isolated(pgdat, file, sc))) {
>>   		if (stalled)
>>   			return 0;
>>   
>> +		if (!is_sc_tmi(sc, file)) {
>> +			set_sc_tmi(sc, file, 1);
>> +			return 0;
>> +		}
>> +
>> +		if (!drained) {
>> +			drained = true;
>> +			lru_add_drain_all();
>> +			continue;
>> +		}
>> +
>>   		/* wait a bit for the reclaimer. */
>>   		msleep(100);
>>   		stalled = true;
>> @@ -2123,6 +2154,9 @@ shrink_inactive_list(unsigned long nr_to
>>   			return SWAP_CLUSTER_MAX;
>>   	}
>>   
>> +	if (is_sc_tmi(sc, file))
>> +		set_sc_tmi(sc, file, 0);
>> +
>>   	lru_add_drain();
>>   
>>   	spin_lock_irq(&lruvec->lru_lock);
>> 
>
>I tried the patch, it still can reproduce the 100ms sleep.
>
>52) @ 103829.8 us |      shrink_lruvec();

Thanks for you data.

What we learn from it is a 5ms nap a time and no longer than 100ms in total
is an acceptable option.

Hillf