Re: [Gluster-users] bonding question
Ok, I mean this is a network based solution, but I think the 100MB/sec is possible with one nic too. I just wondering, maybe my bonding isn't working fine. You should test with multiple clients/dd streams. http://serverfault.com/questions/569060/link-aggregation-lacp-802-3ad-max-throughput/ rr ___ Gluster-users mailing list Gluster-users@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-users
Re: [Gluster-users] Glusterfs high memory usage
PID USER PR NI VIRT RES SHR S %CPU %MEMTIME+ COMMAND 1892 root 20 0 10.2g 4.7g 1900 S 15 61.1 8980:27 glusterfs 10 GBytes is to much for a process and we want to know if there is anything we can do too solve this situation. In general VIRT (virtual) is not the memory you should be looking at but the RES (resident) which is 4.7G and is the actual memory the process is using. You can cleary see it from: Mem: 8125496k total, 5948512k used, 2176984k free Eg your server has only 8Gb of physical ram and 2Gb free so there is no way a process can eat up "actual" 10 Gigs (unless Linux virtual memory). The bad thing though you have quite a large swap size and it's being used: Swap: 14878048k total, 5731656k used, 9146392k free, 109640k cached I would suggest to run: sysctl -w vm.swappiness=0 (and then set it in sysctl.conf for permanet use) so the systems avoids swapping as much as possible. p.s. to understand more about Linux memory management you might to look at this presentation https://www.youtube.com/watch?v=twQKAoq2OPE rr ___ Gluster-users mailing list Gluster-users@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-users
Re: [Gluster-users] Reading directly from brick
Simple answer - no, it's not ever safe to do writes to an active Gluster backend. Question was about reads though and then the answer is it is perfectly fine (and faster) to do reads directly from the filesystem (in replicated setups) if you keep in mind that by doing so you lose the Glusters autoheal eature - eg if one of the gluster nodes goes down and there is a file written meanwhile when the server comes up if you access the file directly it won't show up while it would when accessing it via the gluster mount point (you can work arround it by manually triggering the self heal). I've heard that reads from glusterfs are around 20 times slower than from ext3: "20 times" might be fetched out of thin air but of course there is a significant overhead of serving a file from a gluster which basically involves network operations and additional meta data checks versus fetching the file directly from iron. rr ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
[Gluster-users] Stripe+replicate
Hello, is this ( http://gluster.org/pipermail/gluster-users/2011-July/008223.html ) true regarding 3.3.0 beta or should check out GIT? Also while it is possible to manually create in client volfile will some more complex setups like striped+replicated+distributed setups ( like for example stripe on 6 (or more) nodes each stripe having 3 replicas and distributed on 12 servers) be supported or better stay away from something like that? What's the suggested way to store large ~500 Gb files in reliable way so it doesn't bring the cluster down if a replica fails and has to be resynced. thx in advance rr ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
[Gluster-users] Replacing a downed brick
Hello, while playing around with the new elastic glusterfs system (via 'glusterd', previously have been using glusterfs with static configuration) I have stumbled upon such problem: 1. I have a test system with 12 nodes in a replicated/distributed way (replica count 3): Volume Name: storage Type: Distributed-Replicate Status: Started Number of Bricks: 4 x 3 = 12 Transport-type: tcp 2. One of the brick systems/servers had a simulated hardware failure (disks have been wiped) and restarted a new. 3. When the server ('glusterd') came up the rest of the bricks received something like: Jul 25 17:10:45 snode182 GlusterFS[3371]: [2011-07-25 17:10:45.435786] C [glusterd-rpc-ops.c:748:glusterd3_1_cluster_lock_cbk] 0-: Lock response received from unknown peer: 4ecec354-1d02-4709-8f1e-607a735dbe62 Obviously the peer UID in glusterd.info (because of the full "crash/reinstall") seems to be different from the UID which is in the cluster configuration. Peer status shows: Hostname: 10.0.0.149 Uuid: f9ea651e-68da-40fa-80d9-6bee7779aa97 State: Peer Rejected (Connected) 4. While the info commands work fine anything that involves changing the volume settings return that the volume doesn't exist (from the logs seem to be coming from the reinstalled node): [2011-07-25 17:08:54.579631] E [glusterd-op-sm.c:1237:glusterd_op_stage_set_volume] 0-: Volume storage does not exist [2011-07-25 17:08:54.579769] E [glusterd-op-sm.c:7107:glusterd_op_ac_stage_op] 0-: Validate failed: -1 So my question is how to correctly reintroduce the box to the glusterfs cluster since: 1. I can't run 'peer probe 10.0.0.149' as gluster says the peer is already in cluster 2. I can't remove the peer because it is part of a volume. 3. I can't remove the brick from the volume because gluster asks me to remove 3 bricks (eg the replica count and also would mean data loss). 4. I imagine that the replace-brick won't work even if I fake the new node with a different ip/hostname (since the source brick will be down) or will it replicate from the alive ones? I tried just to manually change the UID back to the the one which is listed in the rest of the nodes (peer status) but apparently it was not enough (the node itself didn't see any other servers and wasn't able to sync volume information from remote brick(s) complaining that it is not his friend). Then when I manually placed all the peers/* files from a running bricknode and restarted glusterd the node reverted to 'Peer in Cluster' state. Is this the way? Or am I doing something totally wrong? wbr rr ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
Re: [Gluster-users] Reading directly from gluster folder ?
I'm curious, however - in what way would you like to use this method? It is useful if you have a replica/mirror setup without distribution then for example you can set up webserver(s) on the node(s) itself and serve directly from filesystem which is faster than through a fuse mountpoint (of course there are drawbacks regarding self-heal but you can work around that too). rr ___ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users