 |
 |
On 3/2/18 1:56 AM, Fabrizio Furano wrote: > Hi Andy, > > yes, I have spent the last few years on various caches and I see the pain. > > For the use cases that I know about the restriction that you mention would be > more than acceptable, and the "total" solution would probably have limited > usefulness (and a high development cost). > Also not caching (hence just tunnelling) a file that is opened for write would > be fine. At the end most of the heavy use cases are just WORM, and whoever > enables this is supposed to be aware of the limitations... hopefully. Do you need tunneling or is redirection good enough? \m > Also limiting the prohibition to opens for append/mpodify would be fine I guess. > > Fabrizio > > > > > > On 03/02/2018 10:46 AM, Andrew Hanushevsky wrote: >> Hi Fabrizio, >> >> The reason we didn't do it for a disk caching proxy is because cache synchrnization is extremely difficult as you may be reading >> as well as writing the file. The memory caching proxy does support reading and writing as synchnizing a memory cache is trivial. >> So, even if we did a passthrough option it would be very restrictive (i.e. you can either read the data or write the data but >> not both unless the cache is disabled and opening a file for writing would necessarily have to flush the cache for that file). >> That we could also do. >> >> Probably the fastest solution here (at the moment) is to write an ofs plugin wrapper for the caching proxy that checks for write >> opens and does the appropriate redirection. We can do that also by extending the static redirect directive to differentiate >> between read paths and write paths; thuogh that opens the proxy to cache synchronization issues unless files are not bth read >> and written at the same time. >> >> Andy >> >> On Fri, 2 Mar 2018, Fabrizio Furano wrote: >> >>> Hi Andy, >>> >>> thanks for the hints. My opinion is that a passthrough option would be >>> a very big plus for the caching proxy, and I was sincerely surprised >>> by this limitation. >>> >>> For some of our use cases I believe that we will start using it without >>> caching, as I can't imagine how to give a transparent service by having >>> to force clients to use different paths for reading and writing. >>> >>> What do you think ? >>> >>> Thank you >>> Fabrizio >>> >>> On 03/01/2018 06:09 PM, Andrew Hanushevsky wrote: >>>> Correct, we don't support a writable cache because that's not what a cache is all about. Does the origin support writes? If so, >>>> we could potentially provide a cache passthrough option. Otherwise, if the readable and writable path differ, he could have run >>>> two proxies -- one caching for read access and a passthru proxy for writing then setup a redirect directive in the primary proxy >>>> (i.e. the one that will be initially used) to direct writes to the passthru proxy based on path. >>>> >>>> Andy >>>> >>>> On Thu, 1 Mar 2018, Fabrizio Furano wrote: >>>> >>>>> Hi, >>>>> >>>>> Mihai is trying to setup a small Pss machine that also works as a cache. >>>>> >>>>> For reading it seems to work fine, yet it stubbornly refuses to write files with >>>>> this error: >>>>> >>>>>> 180301 17:27:38 13788 XrdFileCache_Manager: debug Cache::Attach() >>>>>> root:[log in to unmask]:1094//eos/xdc/testing/hello.txt?&oss.asize=14&oss.lcl=1 location: <deferred open> >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 ofs_fstat: fn=/eos/xdc/testing/hello.txt >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 XrootdResponse: 0100 sending 46 data bytes; status=0 >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 XrootdProtocol: 0100 req=write dlen=14 >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 XrootdProtocol: 0100 fh=0 write 14@0 >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 ofs_write: 14@0 fn=/eos/xdc/testing/hello.txt >>>>>> 180301 17:27:38 13788 ofs_write: root.3008:20@xdc-test-fst1 Unable to write /eos/xdc/testing/hello.txt; operation not >>>>>> supported >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 XrootdProtocol: 0100 discarding 0 bytes >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 XrootdResponse: 0100 sending err 3005: Unable to write >>>>>> /eos/xdc/testing/hello.txt; operation not supported >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 XrootdProtocol: 0100 req=close dlen=0 >>>>>> 180301 17:27:38 13788 root.3008:20@xdc-test-fst1 ofs_close: use=1 fn=/eos/xdc/testing/hello.txt >>>>> >>>>> If we remove the directive pss.cachelib then writes work instead. >>>>> >>>>> Can anyone give us a clue please ? >>>>> >>>>> Thank you >>>>> Fabrizio and Mihai >>>>> >>>>> >>>> >>> >>> > > > ######################################################################## > 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