I have this problem too. My mon's in 0.48.1 cluster have 10GB RAM each, with 78 osd, and 2k request per minute (max) in radosgw. Now i have run one via valgrind. I will send output when mon grow up. On Fri, Aug 31, 2012 at 6:03 PM, Sage Weil <sage@xxxxxxxxxxx> wrote: > On Fri, 31 Aug 2012, Xiaopong Tran wrote: > >> Hi, >> >> Is there any known memory issue with mon? We have 3 mons running, and >> on keeps on crashing after 2 or 3 days, and I think it's because mon >> sucks up all memory. >> >> Here's mon after starting for 10 minutes: >> >> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND >> 13700 root 20 0 163m 32m 3712 S 4.3 0.1 0:05.15 ceph-mon >> 2595 root 20 0 1672m 523m 0 S 1.7 1.6 954:33.56 ceph-osd >> 1941 root 20 0 1292m 220m 0 S 0.7 0.7 946:40.69 ceph-osd >> 2316 root 20 0 1169m 198m 0 S 0.7 0.6 420:26.74 ceph-osd >> 2395 root 20 0 1149m 184m 0 S 0.7 0.6 364:29.08 ceph-osd >> 2487 root 20 0 1354m 373m 0 S 0.7 1.2 401:13.97 ceph-osd >> 235 root 20 0 0 0 0 S 0.3 0.0 0:37.68 kworker/4:1 >> 1304 root 20 0 0 0 0 S 0.3 0.0 0:00.16 jbd2/sda3-8 >> 1327 root 20 0 0 0 0 S 0.3 0.0 13:07.00 xfsaild/sdf1 >> 2011 root 20 0 1240m 177m 0 S 0.3 0.6 411:52.91 ceph-osd >> 2153 root 20 0 1095m 166m 0 S 0.3 0.5 370:56.01 ceph-osd >> 2725 root 20 0 1214m 186m 0 S 0.3 0.6 378:16.59 ceph-osd >> >> Here's the memory situation of mon on another machine, after mon has >> been running for 3 hours: >> >> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND >> 1716 root 20 0 1923m 1.6g 4028 S 7.6 5.2 8:45.82 ceph-mon >> 1923 root 20 0 774m 138m 5052 S 0.7 0.4 1:28.56 ceph-osd >> 2114 root 20 0 836m 143m 4864 S 0.7 0.4 1:20.14 ceph-osd >> 2304 root 20 0 863m 176m 4988 S 0.7 0.5 1:13.30 ceph-osd >> 2578 root 20 0 823m 150m 5056 S 0.7 0.5 1:24.55 ceph-osd >> 2781 root 20 0 819m 131m 4900 S 0.7 0.4 1:12.14 ceph-osd >> 2995 root 20 0 863m 179m 5024 S 0.7 0.6 1:41.96 ceph-osd >> 3474 root 20 0 888m 208m 5608 S 0.7 0.6 7:08.08 ceph-osd >> 1228 root 20 0 0 0 0 S 0.3 0.0 0:07.01 jbd2/sda3-8 >> 1853 root 20 0 859m 176m 4820 S 0.3 0.5 1:17.01 ceph-osd >> 3373 root 20 0 789m 118m 4916 S 0.3 0.4 1:06.26 ceph-osd >> >> And here is the situation on a third node, mon has been running >> for over a week: >> >> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND >> 1717 root 20 0 68.8g 26g 2044 S 91.5 84.1 9220:40 ceph-mon >> 1986 root 20 0 1281m 226m 0 S 1.7 0.7 1225:28 ceph-osd >> 2196 root 20 0 1501m 538m 0 S 1.0 1.7 1221:54 ceph-osd >> 2266 root 20 0 1121m 176m 0 S 0.7 0.5 399:23.70 ceph-osd >> 2056 root 20 0 1072m 167m 0 S 0.3 0.5 403:49.76 ceph-osd >> 2126 root 20 0 1412m 458m 0 S 0.3 1.4 1215:48 ceph-osd >> 2337 root 20 0 1128m 188m 0 S 0.3 0.6 408:31.88 ceph-osd >> >> So, after a while, sooner or later, mon is going to crash, just >> a matter of time. >> >> Does anyone see anything like this? This is kinda scary. >> >> OS: Debian Wheezy 3.2.0-3-amd64 >> Ceph: 0.48argonaut (commit:c2b20ca74249892c8e5e40c12aa14446a2bf2030) > > Can you try with 0.48.1argonaut? > > If it still happens, can you run ceph-mon through massif? > > valgrind --tool=massif ceph-mon -i whatever > > That'll generate a massif.out file (make sure it's there; you may need to > specify the output file for valgrind) over time. Once ceph-mon starts > eating ram, send us a copy of the file and we can hopefully see what is > leaking. > > Thanks! > sage > > >> >> With this issue on hand, I'll have to monitor it closely and >> restart mon once in a while, or I will get a crash (which is >> still good enough), or a system that does not respond at >> all because memory is exhausted, and the whole ceph cluster >> is unreachable. We had this problem in the morning, mon on one >> node exhausted the memory, none of the ceph command responds >> anymore, the only thing left to do is to hard reset the node. >> The whole cluster was basically done at that time. >> >> Here is our usage situation: >> >> 1) A few applications which read and write data through >> librados API, we have about 20-30 connections at any one time. >> So far, our apps have no such memory issue, we have been >> monitoring them closely. >> >> 2) We have a few scripts which pull data from an old storage >> system, and use the rados command to put it into ceph. >> Basically, just shell script. Each rados command is run >> to write one object (one file), and exit. We run about >> 25 scripts simultaneously, which means at any one time, >> there are at most 25 connections. >> >> I don't think this is a very busy system. But this >> memory issue is definitely a problem for us. >> >> Thanks for helping. >> >> Xiaopong >> -- >> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in >> the body of a message to majordomo@xxxxxxxxxxxxxxx >> More majordomo info at http://vger.kernel.org/majordomo-info.html >> >> > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- ----- Pozdrawiam Sławek "sZiBis" Skowron -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
