Perl 6 - the future is here, just unevenly distributed

IRC log for #gluster, 2016-07-08

| Channels | #gluster index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
00:09 hackman joined #gluster
00:18 pdrakeweb joined #gluster
00:41 guhcampos joined #gluster
00:42 guhcampos_ joined #gluster
00:59 ahino joined #gluster
01:07 shdeng joined #gluster
01:19 sage joined #gluster
01:51 cliluw joined #gluster
02:10 shyam joined #gluster
02:14 kramdoss_ joined #gluster
02:19 harish joined #gluster
02:20 fcoelho joined #gluster
02:24 nbalacha joined #gluster
03:34 skoduri joined #gluster
04:06 RameshN joined #gluster
04:09 itisravi joined #gluster
04:11 atinm joined #gluster
04:29 shubhendu joined #gluster
04:31 shubhendu joined #gluster
04:31 nehar joined #gluster
04:33 aspandey joined #gluster
04:35 sakshi joined #gluster
04:47 kdhananjay joined #gluster
04:48 gowtham joined #gluster
04:51 ashiq joined #gluster
04:53 karthik___ joined #gluster
04:54 nishanth joined #gluster
05:01 ndarshan joined #gluster
05:10 Apeksha joined #gluster
05:10 Apeksha joined #gluster
05:12 poornimag joined #gluster
05:19 Muthu__ joined #gluster
05:19 Muthu_ joined #gluster
05:23 sanoj joined #gluster
05:27 prasanth joined #gluster
05:28 Apeksha_ joined #gluster
05:30 karnan joined #gluster
05:40 Lee1092 joined #gluster
05:40 JonathanD joined #gluster
05:51 hchiramm joined #gluster
05:53 nehar joined #gluster
05:53 skoduri joined #gluster
05:56 devyani7_ joined #gluster
06:02 prasanth joined #gluster
06:04 hackman joined #gluster
06:08 jwd joined #gluster
06:09 Apeksha joined #gluster
06:12 satya4ever joined #gluster
06:17 kshlm joined #gluster
06:18 pur joined #gluster
06:20 anil joined #gluster
06:21 nbalacha joined #gluster
06:23 jtux joined #gluster
06:28 atalur joined #gluster
06:28 post-factum masuberu: how those files are organized?
06:30 jiffin joined #gluster
06:32 masuberu in folders of 93 files
06:32 masuberu no more than 6 level folders
06:33 ramky joined #gluster
06:33 Pupeno joined #gluster
06:34 masuberu post-factum: not sure if I answered your question
06:34 post-factum masuberu: that is what i wanted to hear
06:35 post-factum masuberu: also, is there some index of those files in some database?
06:35 masuberu no
06:35 masuberu so the scenario would be
06:36 masuberu I have 10 machines that creates 1.2 millions files (around 800 Gig) each
06:36 masuberu while they are doing one job
06:36 masuberu this jobs takes 3 days
06:36 masuberu one the job finishes they start again
06:36 masuberu it is like a farm
06:37 masuberu all those files would land on gluster
06:37 masuberu throught network
06:37 masuberu SAMBA because those 10 machines are windows 7
06:37 post-factum ok, how those millions of files would be used further?
06:38 masuberu that is the easy part because there is no much IOPs
06:38 masuberu then
06:38 masuberu I have another cluster
06:38 masuberu linux
06:38 masuberu which would be monitoring the glusterFS cluster
06:38 Alghost joined #gluster
06:39 masuberu looking for "jobs that have finished"
06:39 post-factum masuberu: iow, scanning the whole bunch of folders?
06:39 masuberu each job --> 1.2Million files (800G) inside a folder
06:39 masuberu well
06:40 masuberu to know if the job has finished they just need to look for a specific file
06:40 masuberu inside de job folder
06:40 masuberu no need to scan everything
06:40 post-factum masuberu: ok then
06:41 post-factum should work for you as long as you do not keep many files within one folder and do not scan those millions each time
06:41 shubhendu_ joined #gluster
06:41 post-factum sorting files across subfolders works well
06:41 masuberu but one they find a folder that needs to be processed they they need to copy the whole job folder to local storage of the linux cluster
06:41 masuberu that is the critical part because it will kill the IOPS
06:42 masuberu and at the moment my linux cluster has 6 servers
06:42 post-factum copying lots of small files is indeed slow
06:42 masuberu ok
06:43 masuberu well copying small files are slow everywhere
06:43 masuberu I just needs to know if this is a disadvantage for gluster
06:43 post-factum this is a disadvantage of each network filesystem
06:43 masuberu correct
06:44 post-factum because each file operation time is multiplied by RTT
06:46 masuberu I was wondering if the fact of using files spread across multiple hosts would give any advantage
06:46 Saravanakmr joined #gluster
06:47 masuberu I just need replication not distribution
06:47 masuberu as my files are small no need to split them
06:47 rastar joined #gluster
06:48 post-factum replication slows things down, but if you need redundancy and high availability, it should be used
06:49 masuberu humm I see
06:50 Klas just curious, is it more of an archive or are the files often changed?
06:51 Klas if you mostly download them, you could put them into tarballs (not even zipped) to improve handling of them
06:52 Klas has definitie disadvantages as well of course, since then you need to handle the files as a repository which needs to uploaded/downloaded in full
06:53 masuberu the files are not modified
06:53 masuberu they are copied to another place
06:53 masuberu and then deleted when they are not needed
06:54 Klas then it would probably go faster to put all grouped files into a tarball and then unpacking the tarball on destination
06:54 Klas (as long as you can handle double to disk space
06:54 Klas )
06:54 Klas temporarily
06:54 masuberu mmm
06:54 Klas then you get to avoid the "stat every file" aspect
06:55 masuberu but I need to read of one of the files to know that the job folder is ready
06:55 masuberu and it will take long time to tar then
06:55 Klas hmm, uncertain if you can read a file in a tarball in a simple way
06:56 masuberu I can't wait to tar the files, once one folder is ready I need to take it to my analytic cluster to process the data
06:57 Klas ah, ok then, then a tar archive is meaningless
06:57 Klas it will take as long to create a tarball as copying them
06:57 Klas give or take
07:06 shubhendu_ joined #gluster
07:10 kramdoss_ joined #gluster
07:16 deniszh joined #gluster
07:26 arcolife joined #gluster
07:27 jri joined #gluster
07:29 Pupeno joined #gluster
07:34 [Enrico] joined #gluster
07:40 kdhananjay Ulrar: There?
07:41 Ulrar kdhananjay: Yes
07:42 Ulrar kdhananjay: So about my mail from yesterday, I removed the brick with the dead raid card, changed it then added that brick again
07:42 Ulrar It started a big "heal" of 2500 shards since it was a add-brick, no freez at all
07:42 Ulrar We'll still need to test the heal in a real condition, but at least for an add brick it doesn't freez so that's a good sign I think
07:43 karthik___ joined #gluster
07:43 Ulrar I'm guessing the faulty hardware was what caused the freez, does that seem possible ?
07:44 prasanth joined #gluster
07:46 hackman joined #gluster
07:48 kdhananjay Ulrar: im not sure i understood, are you saying it is working fine now?
07:48 Ulrar kdhananjay: Well when I added the brick increasing the replica to 3, it started a heal of all the shards
07:48 Ulrar And that worked fine
07:49 Ulrar Did it in the middle of the afternoon and no-one noticed
07:49 kdhananjay Ulrar: yeah, add-brick will heal the whole directory tree of the volume from the existing replicas to the new replica
07:49 kdhananjay Ulrar: what was the size of your volume at that time?
07:49 Klas Ulrar: you mean no-one complained =P?
07:49 Ulrar Klas: I guess haha
07:50 Ulrar kdhananjay: About 160 Gb
07:50 Ulrar With 64 Mb shards it was about 2500 files to heal
07:50 Klas Ulrar: I just assumed you didn't perform a survey ;)
07:50 Ulrar Klas: Our monitoring didn't start to complain
07:50 nbalacha joined #gluster
07:51 kdhananjay Ulrar: hmm that *may* have increased the load on the volume.
07:51 Klas Ulrar: good =)
07:52 Ulrar Well, it worked perfectly anyway. We will kill a node one of these nights to test an actual heal and not an add brick
07:52 Ulrar Just to be sure
07:52 Ulrar I'm hoping to do that next week
07:52 kshlm joined #gluster
07:53 kdhananjay Ulrar: ah, ok. so do i treat your issue on gluster users as resolved?
07:54 Ulrar kdhananjay: For now, sure. If the test next week doesn't go well I'll write a new mail
07:54 Ulrar thanks !
07:55 kdhananjay Ulrar: ok.
08:00 Pupeno joined #gluster
08:02 Pupeno joined #gluster
08:02 hybrid512 joined #gluster
08:07 Slashman joined #gluster
08:10 Pupeno joined #gluster
08:11 ivan_rossi joined #gluster
08:16 nage joined #gluster
08:16 Pupeno joined #gluster
08:21 Pupeno joined #gluster
08:25 aravindavk joined #gluster
08:26 Pupeno joined #gluster
08:29 Seth_Karlo joined #gluster
08:37 test12 joined #gluster
08:38 test12 gluster
08:43 ramky joined #gluster
08:44 Wizek joined #gluster
09:06 Pupeno joined #gluster
09:11 Pupeno joined #gluster
09:13 Peppard joined #gluster
09:17 kramdoss_ joined #gluster
09:31 jiffin1 joined #gluster
09:38 kramdoss_ joined #gluster
09:48 PaulCuzner joined #gluster
09:56 kdhananjay1 joined #gluster
09:57 sakshi joined #gluster
09:59 jiffin1 joined #gluster
10:00 Saravanakmr joined #gluster
10:01 [Enrico] joined #gluster
10:03 ira joined #gluster
10:05 AppStore Which would yield best latency performance for stat-ing a file, GlusterFS FUSE Client, or NFS Ganesha + FUSE client + NFS Client + NFS cache?
10:05 kdhananjay joined #gluster
10:38 ndevos AppStore: why not NFS-Ganesha + FSAL_GLUSTER ?
10:38 AppStore Whichever lets me serve a WordPress site.. :)
10:39 AppStore With reasonable performance.
10:39 AppStore Last time I tried this was around GlusterFS 3.5 and performance using FUSE back then was bad.
10:39 AppStore I understand a lot of things has changed since then though, so looking into it again.
10:42 ndevos AppStore: the nfs-client will do some caching of stat attributes, no communicatino with the NFS-server makes that fastest :)
10:42 ndevos also, ,,(php) might help
10:42 glusterbot (#1) php calls the stat() system call for every include. This triggers a self-heal check which makes most php software slow as they include hundreds of small files. See http://joejulian.name/blog/optimizing-web-performance-with-glusterfs/ for details., or (#2) It could also be worth mounting fuse with glusterfs --attribute-timeout=HIGH --entry-timeout=HIGH --negative-timeout=HIGH --fopen-keep-cache
10:43 ndevos AppStore: you can make the NFS-client cache longer by passing some of the ac* mount options, see 'man 5 nfs' for details
11:06 jwd joined #gluster
11:06 shyam joined #gluster
11:10 johnmilton joined #gluster
11:12 johnmilton joined #gluster
11:12 aspandey joined #gluster
11:21 aspandey joined #gluster
11:26 PaulCuzner joined #gluster
11:28 wadeholler joined #gluster
11:38 wadeholler joined #gluster
12:08 Jacob843 joined #gluster
12:08 chirino joined #gluster
12:20 B21956 joined #gluster
12:20 shyam joined #gluster
12:38 Gnomethrower joined #gluster
12:47 plarsen joined #gluster
13:00 jwd joined #gluster
13:09 unclemarc joined #gluster
13:20 AppStore Hmm, interesting. Using FUSE, find /mywordpressapp takes between 3-4 seconds. Using old NFS, it takes about 0.3-0.4 seconds. Using Ganesha, it takes about 0.2-0.3 seconds.
13:40 squizzi joined #gluster
13:40 bowhunter joined #gluster
13:43 edong23 joined #gluster
13:43 kdhananjay joined #gluster
14:07 jri_ joined #gluster
14:09 jri__ joined #gluster
14:09 hagarth joined #gluster
14:11 Wizek joined #gluster
14:23 unclemarc joined #gluster
14:26 harish_ joined #gluster
14:30 Wizek joined #gluster
14:38 poornimag joined #gluster
14:45 hagarth joined #gluster
15:03 wushudoin joined #gluster
15:10 shaunm joined #gluster
15:14 kpease joined #gluster
15:15 rafaels joined #gluster
15:16 ivan_rossi left #gluster
15:19 level7 joined #gluster
15:24 kpease joined #gluster
15:35 rafaels joined #gluster
15:44 robb_nl joined #gluster
15:49 gnulnx If my brick directory is tank/bricks/ftp and all of my files are under /tank/bricks/ftp/ftp (that is, there is one parent folder under the brick for 1.5 million files totally 30T of files), would that affect the distribution of files in a distributed volume?
15:52 kaushal_ joined #gluster
16:02 dblack joined #gluster
16:20 Manikandan joined #gluster
16:25 shyam joined #gluster
16:32 cloph_away gnulnx: that largely would affect directory listings, don't think so much  distribution, but accessing the files. (if you mwan that all files are in that single subdirectory)
16:44 ira_ joined #gluster
16:45 itisravi|android joined #gluster
16:48 jiffin joined #gluster
16:49 itisravi|android joined #gluster
16:55 hagarth1 joined #gluster
17:01 dnunez joined #gluster
17:05 squizzi_ joined #gluster
17:10 tertiary joined #gluster
17:11 tertiary @JoeJulian, know of any fixes for "Error 0x80070021: The process cannot access the file because another process has locked a portion of the file" when using Samba with "stict locking = yes" from a Windows client?
17:11 tertiary *strict
17:12 Manikandan joined #gluster
17:17 overclk joined #gluster
17:37 skylar joined #gluster
17:42 karnan joined #gluster
17:46 level7_ joined #gluster
17:53 jiffin joined #gluster
18:27 ashka joined #gluster
18:27 ashka joined #gluster
19:06 bowhunter joined #gluster
19:40 shyam joined #gluster
19:49 PaulCuzner joined #gluster
20:11 PaulCuzner joined #gluster
20:18 guhcampos joined #gluster
20:38 arcolife joined #gluster
20:52 shyam joined #gluster
20:56 deniszh joined #gluster
21:12 deniszh joined #gluster
21:48 shaunm joined #gluster
22:03 hackman joined #gluster
22:16 gbox I accidentally deleted files from brick directory.  Gluster volume now has I/O errors but I'm not sure how to fix them.
22:41 gbox Do I just need to remove the orphans in the .glusterfs directory and the afr translator will copy over the files?
22:43 gbox Does a file have the same gfid on each replicated brick?
23:40 plarsen joined #gluster
23:44 hackman joined #gluster
23:59 Gnomethrower joined #gluster

| Channels | #gluster index | Today | | Search | Google Search | Plain-Text | summary