Perl 6 - the future is here, just unevenly distributed

IRC log for #gluster-dev, 2015-11-17

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

All times shown according to UTC.

Time Nick Message
00:03 suliba joined #gluster-dev
00:14 shyam joined #gluster-dev
00:49 amye joined #gluster-dev
01:25 RedW joined #gluster-dev
02:46 mchangir joined #gluster-dev
03:01 kshlm joined #gluster-dev
03:07 pranithk joined #gluster-dev
03:11 pranithk joined #gluster-dev
03:18 msvbhat joined #gluster-dev
03:19 mchangir joined #gluster-dev
03:19 gem joined #gluster-dev
03:39 overclk joined #gluster-dev
03:53 sakshi joined #gluster-dev
03:59 nishanth joined #gluster-dev
04:00 shubhendu joined #gluster-dev
04:01 nbalacha joined #gluster-dev
04:04 dlambrig_ Hi Nithya, are you there?
04:10 atinm joined #gluster-dev
04:11 itisravi joined #gluster-dev
04:11 nbalacha joined #gluster-dev
04:15 hchiramm_home joined #gluster-dev
04:19 itisravi joined #gluster-dev
04:35 zhangjn joined #gluster-dev
04:40 ppai joined #gluster-dev
04:40 nbalacha joined #gluster-dev
04:48 hchiramm_home joined #gluster-dev
04:51 bkunal|training joined #gluster-dev
04:52 josferna joined #gluster-dev
04:56 csim_ joined #gluster-dev
04:56 lalatend1M joined #gluster-dev
04:56 xavih_ joined #gluster-dev
04:58 Humble joined #gluster-dev
05:01 kbyrne joined #gluster-dev
05:07 ndarshan joined #gluster-dev
05:20 suliba_ joined #gluster-dev
05:28 atinm joined #gluster-dev
05:30 mmckeen joined #gluster-dev
05:31 pppp joined #gluster-dev
05:31 aravindavk joined #gluster-dev
05:38 josferna joined #gluster-dev
05:40 Chr1st1an joined #gluster-dev
05:42 atalur joined #gluster-dev
05:47 vimal joined #gluster-dev
05:51 jiffin joined #gluster-dev
05:58 rafi joined #gluster-dev
05:58 Bhaskarakiran joined #gluster-dev
06:03 hgowtham joined #gluster-dev
06:03 hgowtham_ joined #gluster-dev
06:12 vmallika joined #gluster-dev
06:13 bkunal|training joined #gluster-dev
06:20 pranithk joined #gluster-dev
06:20 kdhananjay joined #gluster-dev
06:21 anekkunt joined #gluster-dev
06:21 raghu joined #gluster-dev
06:29 pranithk joined #gluster-dev
06:30 Saravana_ joined #gluster-dev
06:35 jiffin1 joined #gluster-dev
06:38 mchangir joined #gluster-dev
06:40 asengupt joined #gluster-dev
06:41 jiffin1 joined #gluster-dev
06:57 aravindavk joined #gluster-dev
07:15 deepakcs joined #gluster-dev
07:31 pranithk joined #gluster-dev
07:50 pranithk joined #gluster-dev
08:11 suliba joined #gluster-dev
08:17 suliba joined #gluster-dev
08:34 kanagaraj joined #gluster-dev
08:36 poornimag joined #gluster-dev
08:37 josferna joined #gluster-dev
08:43 suliba joined #gluster-dev
08:44 deepakcs joined #gluster-dev
09:04 deepakcs joined #gluster-dev
09:11 pranithk xavih: hey, josferna and I were talking about tiering + ec related concern. Could you please ping when you are available so that we can complete that conversation?
09:13 xavih pranithk: I'm just working for a customer. Let me an hour or so and I'll be available
09:15 deepakcs joined #gluster-dev
09:18 ndevos rafi, kdhananjay, jiffin, raghu: could one of you host the bug triage meeting later today?
09:18 pranithk xavih: sure. Ping once you are available :-)
09:18 pranithk josferna: ^^
09:21 atinm joined #gluster-dev
09:21 kanagaraj joined #gluster-dev
09:23 pranithk xavih: josferna: FYI: We just fixed an important bug with tiering + disperse. http://review.gluster.com/12595 thanks to atalur and rafi
09:26 suliba joined #gluster-dev
09:28 kshlm joined #gluster-dev
09:54 suliba joined #gluster-dev
10:21 jvava joined #gluster-dev
10:23 pppp joined #gluster-dev
10:24 kanagaraj_ joined #gluster-dev
10:28 overclk joined #gluster-dev
10:31 xavih pranithk: ping. I'm done
10:33 pranithk xavih: josferna brought up an interesting detail about how the way ec/afr do internal read/write operations to the files which are "heating up" the files when self-heal is in progress. Similarly in ec we have read-modify-write when the write falls in between the chunks
10:34 pranithk xavih: These internal fops shouldn't heat up the files. So for healing we already have -ve pid for the fops. So we can filter them out. But for read-modify-write, we don't have such a hook to identify that these operations are internal
10:35 pranithk xavih: the question is, how should we treat these reads? josferna, tier is supposed to differentiate between application vs internal xlator read/writes is it?
10:36 xavih pranithk: currently an anonymous fd is used for those reads, but I suppose this is not enough to differentiate them from normal reads from NFS
10:36 pranithk xavih: that is true :-/
10:37 xavih pranithk: maybe those reads could be also done with a negative pid
10:37 josferna pranithk: yep ... but without the marker on the fops we cannot differentiate normal fops from internal fops
10:38 josferna xavih: yep a marker on reads or any fops that EC thinks is internal
10:39 xavih pranithk, josferna: I think the negative pid for internal fops should be easy to implement
10:40 pranithk xavih: I agree!
10:40 pranithk josferna: what do you think?
10:41 pranithk xavih: wait
10:41 josferna xavih, pranithk : I think that should be sufficient ...
10:42 josferna xavih, pranithk : if you could send a patch I would test it
10:42 pranithk xavih: I don't think anon-fd pid is sent over network... wait
10:42 xavih pranithk: that would be a problem...
10:42 suliba joined #gluster-dev
10:48 pranithk xavih: I think we are already changing uid/gid to '0' we can set it to -1 and reset it at the same place?
10:48 pranithk xavih: I mean we can give it self-heal pid
10:49 xavih pranithk: yes. That's what I meant, but we need to be sure that this also works for anonymous fd
10:49 suliba joined #gluster-dev
10:49 pranithk xavih: we can set frame->root->pid -ve, nothing to do with fd
10:49 xavih pranithk: ok, then the patch should be easy :)
10:50 xavih pranithk: I can write it if you want
10:50 pranithk xavih: I found one bug though... we are changing it back to the previous value in the same function, which is wrong IMO, in case readv fails on one of the bricks and it tries to read the next brick....
10:51 xavih pranithk: I think you are right. Let me check
10:51 pranithk xavih: Did you get what I am saying? because ec tries to wind  to 'min' bricks. If some of those fops fail it falls back to the next set of bricks. But by that time the uid/gid values will be different.
10:52 xavih pranithk: yes. This also need to be solved
10:53 xavih pranithk: I'll write a patch for both things
10:53 xavih pranithk: is there any bug if for this ?
10:53 pranithk xavih: cool. I think we should have fop->uid/gid/pid which we set at the beginning of the fop from frame->uid/gid/pid. Whenever the internal fop completes if the frame->uid/gid/pid mismatch with fop->uid/gid/pid.
10:53 xavih pranithk: *bug id*
10:53 pranithk josferna: ^^
10:55 pranithk xavih: meanwhile, do you agree with the solution?
10:56 josferna xavih, pranithk : https://bugzilla.redhat.co​m/show_bug.cgi?id=1278279
10:56 glusterbot Bug 1278279: unspecified, unspecified, ---, josferna, ASSIGNED , EC: File healing promotes it to hot tier
10:56 xavih pranithk: yes. I think it's ok. I'll think about the details when I write the patch
10:56 pranithk josferna: that is redhat internal product bug. Is there upstream bug?
10:57 pranithk xavih: cool! thanks.
10:58 josferna pranithk, xavih : cloned one https://bugzilla.redhat.co​m/show_bug.cgi?id=1282761
10:58 glusterbot Bug 1282761: unspecified, unspecified, ---, bugs, NEW , EC: File healing promotes it to hot tier
10:59 josferna xavih, pranithk : thanks guys :)
11:00 atinm joined #gluster-dev
11:00 xavih josferna: np :)
11:01 pranithk xavih: Completely unrelated question. How are the changes for 3.8 going?
11:03 pranithk xavih: Do let me know if you need more people for anything.... I will be happy to help out
11:03 pranithk xavih: I will also understand the algos this time when I help out ;-). That is added bonus :-)
11:04 xavih pranithk: they are a bit delayed, but will work on them very soon. However I'll need to discuss an important topic... :-/
11:04 kkeithley1 joined #gluster-dev
11:05 pranithk xavih: tell me
11:06 xavih pranithk: what do you think about using dynamic code ?
11:06 pranithk xavih: oh, what does it mean? :-(
11:07 xavih pranithk: to generate small pieces of binary code on execution time to do the multiplications
11:09 xavih pranithk: the benefit is that it's *much* faster. Jerasure using SSE is almost 10 times slower than the dynamic method using only 64 bit instructions (I still have to test it with SSE/AVX)
11:09 pranithk xavih: yeah, I remember a mail you sent to one of my colleagues giving that. I am not very knowledgeable in this area. If you feel that is the best way forward lets go that way. I will meantime learn what it is about...
11:09 xavih pranithk: though I've to say that I'm not an expert on jerasure and I'm not sure I'm using the fastest option
11:11 xavih pranithk: I'll post a patch for this very soon so we can discuss it
11:11 pranithk xavih: We can probably find out if there is anyone in Redhat who knows more about jerasure.
11:12 pranithk dlambrig_: ^^ ceph uses jerasure right? Do we have any contact who knows more about jerasure?
11:12 xavih pranithk: that would be useful. I've read the code and used something very similar to what one of the examples does to implement reed-solomon
11:12 zhangjn joined #gluster-dev
11:12 xavih pranithk: but it seems to be very slow for small block sizes
11:12 xavih pranithk: maybe there's a way to prepare the computation to avoid those delays on small blocks
11:13 pranithk xavih: dan i.e. dlambrig_ knows people from ceph team who implemented erasure coding. So may be we can pick his brain. Let me see if we can get him on #gluster-dev sometime.
11:13 zhangjn joined #gluster-dev
11:14 nbalacha joined #gluster-dev
11:14 zhangjn joined #gluster-dev
11:15 xavih pranithk: great. Thanks :)
11:15 pranithk xavih: Let me send a mail right away CCing you to Dan
11:15 zhangjn joined #gluster-dev
11:16 zhangjn joined #gluster-dev
11:17 zhangjn joined #gluster-dev
11:18 pranithk xavih: done.
11:18 zhangjn joined #gluster-dev
11:20 zhangjn joined #gluster-dev
11:21 zhangjn joined #gluster-dev
11:23 pranithk xavih: there is a problem with doing reads with anon-fds :-(
11:24 pranithk xavih: If the file is deleted then we have a problem...
11:24 pranithk xavih: Sorry I must have found these problems earlier. But better late than never :-/
11:26 pppp joined #gluster-dev
11:30 xavih pranithk: I think we already talked about that some time ago (or I talked to someone else). The problem using the same fd is that it fails when the fd is opened for RO
11:31 anekkunt joined #gluster-dev
11:31 xavih pranithk: we could change the flags so that the files are always opened for RW and check for RO inside ec, but this has the problem that trying to open a file with RO access rights (r--) fails
11:31 glusterbot xavih: (r's karma is now -1
11:32 xavih pranithk: I don't see any clean solution for this that also considers deleted files
11:33 pranithk xavih: hmm... interesting problem. Let me also think about this.
11:37 pranithk xavih: For the earlier problem, I saw that at one place we are not setting the self-heal pid for self-heal. I also changed the name of the enum to reflect that EC is also using it. I sent a patch @ http://review.gluster.org/12598 most probably this shouldn't conflict with the patch you are going to send. Do rebase in case it conflicts :-)
11:42 rjoseph joined #gluster-dev
11:52 ndevos REMINDER: Gluster Bug Triage meeting starts in 10 minutes in #gluster-meeting
11:55 rraja_ joined #gluster-dev
12:02 ndevos REMINDER: Gluster Bug Triage meeting starts *NOW* in #gluster-meeting
12:42 anekkunt joined #gluster-dev
12:52 josferna joined #gluster-dev
12:56 ira joined #gluster-dev
13:08 jeffrin joined #gluster-dev
13:09 jeffrin hello all
13:11 jeffrin anyone here ?
13:14 rjoseph joined #gluster-dev
13:15 sakshi joined #gluster-dev
13:15 jeffrin left #gluster-dev
13:39 bkunal|training joined #gluster-dev
13:46 rjoseph joined #gluster-dev
14:01 shyam joined #gluster-dev
14:27 bfoster joined #gluster-dev
14:29 gem joined #gluster-dev
14:36 nishanth joined #gluster-dev
14:40 jiffin joined #gluster-dev
14:45 nbalacha joined #gluster-dev
14:58 shubhendu joined #gluster-dev
15:24 kdhananjay joined #gluster-dev
16:00 sakshi joined #gluster-dev
16:15 cholcombe joined #gluster-dev
16:36 RedW joined #gluster-dev
17:07 rafi joined #gluster-dev
17:11 overclk joined #gluster-dev
17:59 kkeithley1 joined #gluster-dev
18:09 rraja_ joined #gluster-dev
18:12 Humble joined #gluster-dev
18:21 rafi1 joined #gluster-dev
18:29 Chr1st1an joined #gluster-dev
18:56 lpabon joined #gluster-dev
19:26 Chr1st1an joined #gluster-dev
19:54 kkeithley1 joined #gluster-dev
21:02 wushudoin joined #gluster-dev
21:09 dlambrig_ joined #gluster-dev
21:15 dlambrig_ joined #gluster-dev
21:16 dlambrig_ joined #gluster-dev
22:06 dlambrig_ joined #gluster-dev

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