 |
 |
Hi Nikolai, On 2020-02-27 01:59, Nikolai Hartmann wrote: > Hi Matevz, > > Is this a counter for each file? So if i have pfc.blocksize 1M for each > file every 2000M that are read the cinfo file is written? > So for this reporting script i could (if nothing else goes wrong) only > miss file accesses within the last hour if less than 2000M were read in > the last hour? Not quite, this is only active while writing to a file, and the flush limit applies to data written into the file. When a data file is complete, the cinfo file will only be updated when the file is closed. Matevz > Thanks, > Nikolai > > On 2/26/20 11:13 PM, Matevz Tadel wrote: >> Hi, >> >> It's determined by (apprently undocumented) parameter pfc.flush >> >> pfc.flush [blocks | bytes[kmg]] >> >> The default is 2000 blocks. >> >> At this point data files is fsynced, cinfo file is written/updated and >> also fsynced. >> >> Cheers, >> Matevz >> >> On 2020-02-25 15:19, Andrew Hanushevsky wrote: >>> Hi Nikolai, >>> >>> I'm sure Matevz can asnwer the question. However, I do want you to be >>> aware that cinf files are considered private information. There is no >>> public API defined for them (though monitoring information will be >>> generated from them in the next major release). This means that they >>> may change without warning and at any time. >>> >>> Andy >>> >>> On Tue, 25 Feb 2020, Nikolai Hartmann wrote: >>> >>>> Hi, >>>> >>>> We are using a script from Ilija to report file access statistics from >>>> our xcache server: >>>> >>>> https://github.com/slateci/XCache/blob/master/cacheReporter/reporter.py >>>> >>>> This looks for cinfo files modified in the last hour and then in these >>>> files looks for accesses within the last hour to report. The idea is >>>> then to run this at least once an hour to report all file accesses >>>> (minus the detail that only the last 20 will be reported if more than 20 >>>> accesses happened in the last hour). >>>> >>>> I'm currently trying to debug why i sometimes don't report anything >>>> although i can see when dumping the cinfo files later that at these >>>> times there should have been accesses. >>>> >>>> Are the cinfo files written immediately after file accesses or could it >>>> be that they are written later, such that i miss accessed files with >>>> this method? >>>> >>>> Thanks, >>>> >>>> Nikolai >>>> >>>> ######################################################################## >>>> Use REPLY-ALL to reply to list >>>> >>>> To unsubscribe from the XROOTD-L list, click the following link: >>>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1 >>>> >>> >>> ######################################################################## >>> Use REPLY-ALL to reply to list >>> >>> To unsubscribe from the XROOTD-L list, click the following link: >>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1 >> ######################################################################## Use REPLY-ALL to reply to list To unsubscribe from the XROOTD-L list, click the following link: https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1