Cleanup of secondary proc fbarray files?

Cleanup of secondary proc fbarray files?

[Eads, Gage]

As far as I can tell, DPDK does not destroy secondary process fbarray files - i.e. those whose names end with "_<PID>". With enough secondary processes and memory usage per application, and after enough repeat executions, these can take up a significant amount of space. Is the user expected to clean these up themselves, or is this a bug in DPDK?

Perhaps this is a good candidate for including in rte_eal_cleanup()?

Thanks,
Gage

Re: Cleanup of secondary proc fbarray files?

[Burakov, Anatoly]

On 31-Jul-18 5:36 PM, Eads, Gage wrote:
> As far as I can tell, DPDK does not destroy secondary process fbarray 
> files – i.e. those whose names end with “_<PID>”. With enough secondary 
> processes and memory usage per application, and after enough repeat 
> executions, these can take up a significant amount of space. Is the user 
> expected to clean these up themselves, or is this a bug in DPDK?
> 
> Perhaps this is a good candidate for including in rte_eal_cleanup()?
> 
> Thanks,
> 
> Gage
> 

Good point, this was my omission. This should be done in eal_cleaup().

-- 
Thanks,
Anatoly

Re: Cleanup of secondary proc fbarray files?

[Burakov, Anatoly]

On 01-Aug-18 9:01 AM, Burakov, Anatoly wrote:
> On 31-Jul-18 5:36 PM, Eads, Gage wrote:
>> As far as I can tell, DPDK does not destroy secondary process fbarray 
>> files – i.e. those whose names end with “_<PID>”. With enough 
>> secondary processes and memory usage per application, and after enough 
>> repeat executions, these can take up a significant amount of space. Is 
>> the user expected to clean these up themselves, or is this a bug in DPDK?
>>
>> Perhaps this is a good candidate for including in rte_eal_cleanup()?
>>
>> Thanks,
>>
>> Gage
>>
> 
> Good point, this was my omission. This should be done in eal_cleaup().
> 
Actually, it should probably be done at fbarray allocation :) We put a 
lock on those files, so those that are unlocked we know can be removed 
safely. We do the same for hugetlbfs files at init. I will submit a 
patch for this for 18.11 some time later.

-- 
Thanks,
Anatoly