>>> Digimer <li...@alteeve.ca> schrieb am 15.01.2014 um 19:38 in Nachricht <52d6d5a5.9020...@alteeve.ca>: > On 15/01/14 05:55 AM, Ulrich Windl wrote: >> Hi! >> >> I'm wondering what is going on with SLES11 SP3 and cluster communication: > cLVM with mirroring seems to bring cluster communication down. I had this > problem in SP2, and support told me in SP3 (which was not available at that > time) thing should be better. Now I have SP3, but things aren't better. >> Maybe it's time to take care of the problems; one by one. >> >> The cluster seems to have a longish retransmit list, but the list seems > inconsistent by itself: The same items appear over and over (which would > indicate no transmission is possible), but then some items in the list > change, (which seems to indicating that some transfer must have succeeded). > But looking that the list as a total, I cannot make any sense from it. >> >> See yourself: >> [...] >> Jan 15 08:55:29 o5 corosync[13636]: [TOTEM ] Retransmit List: d252 d254 > d256 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d253 d255 d257 >> Jan 15 08:55:30 o5 corosync[13636]: [TOTEM ] Retransmit List: d253 d255 > d257 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d252 d254 d256 >> Jan 15 08:55:30 o5 corosync[13636]: [TOTEM ] Retransmit List: d252 d254 > d256 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d253 d255 d257 >> Jan 15 08:55:30 o5 corosync[13636]: [TOTEM ] Retransmit List: d253 d255 > d257 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d252 d254 d256 >> Jan 15 08:55:31 o5 corosync[13636]: [TOTEM ] Retransmit List: d252 d254 > d256 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d253 d255 d257 >> Jan 15 08:55:31 o5 corosync[13636]: [TOTEM ] Retransmit List: d253 d255 > d257 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d252 d254 d256 >> Jan 15 08:55:32 o5 corosync[13636]: [TOTEM ] Retransmit List: d252 d254 > d256 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d253 d255 d257 >> Jan 15 08:55:32 o5 corosync[13636]: [TOTEM ] Retransmit List: d253 d255 > d257 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d252 d254 d256 >> Jan 15 08:55:32 o5 corosync[13636]: [TOTEM ] Retransmit List: d252 d254 > d256 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d253 d255 d257 >> Jan 15 08:55:33 o5 corosync[13636]: [TOTEM ] Retransmit List: d253 d255 > d257 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d252 d254 d256 >> Jan 15 08:55:33 o5 corosync[13636]: [TOTEM ] Retransmit List: d252 d254 > d256 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d253 d255 d257 >> Jan 15 08:55:34 o5 corosync[13636]: [TOTEM ] Retransmit List: d253 d255 > d257 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d252 d254 d256 >> Jan 15 08:55:34 o5 corosync[13636]: [TOTEM ] Retransmit List: d252 d254 > d256 d241 d242 d243 d244 d245 d246 d247 d248 d249 d24a d24b d24c d24d d24e > d24f d250 d251 d253 d255 d257 >> [...] >> >> In between these messages I see some device-mapper messages, usure whether > thats the culprit or the victim: >> Jan 15 08:55:11 o5 kernel: [ 758.400071] device-mapper: dm-log-userspace: > [35cRCORE] Request timed out: [15/186129] - retrying >> Jan 15 08:55:13 o5 kernel: [ 760.300067] device-mapper: dm-log-userspace: > [35cRCORE] Request timed out: [9/186130] - retrying >> Jan 15 08:55:28 o5 kernel: [ 775.300067] device-mapper: dm-log-userspace: > [35cRCORE] Request timed out: [9/186132] - retrying >> >> I hope these messages don't try to tell me that there are 186 thousand > requests pending ;-) >> Despite of that the message above could benefit from an actual device name > being included in the messages. >> >> Regards, >> Ulrich > > I can't speak to Suse, but I've seen this with RHEL 6.1 when there was a > short-lived bug caused by the hardware not being fast enough (sorry, the > details are fuzzy), which was fixed. The reason I say this is that, not > knowing the SP2/SP3 issue, my first thought is to look at the network stack. > > Can you elaborate on how 'clvmd with mirroring' triggers this? I've used > clvmd a lot, but I've never looked at mirroring in LVM (though I know > it's possible). > > If possible, can you share your cluster.conf (and crm configure show if > using pacemaker, too)?
The cluster configuration is a mess, because I had to disable most things because of the OCFS problem describes yesterday; anyway I try to present some details: --- # grep -v '^ *$' /etc/corosync/corosync.conf | grep -v '^[ ]*#' aisexec { group: root user: root } service { use_mgmtd: yes use_logd: yes ver: 0 name: pacemaker } totem { rrp_mode: passive window_size: 100 join: 60 max_messages: 20 vsftype: none token: 5000 consensus: 6000 secauth: on token_retransmits_before_loss_const: 10 threads: 4 transport: udp version: 2 interface { bindnetaddr: 172.20.3.0 mcastaddr: 239.192.3.9 mcastport: 5405 ringnumber: 0 } interface { mcastaddr: 239.192.3.109 mcastport: 5405 bindnetaddr: 192.168.0.0 ringnumber: 1 } clear_node_high_bit: yes } logging { to_logfile: no to_syslog: yes debug: off timestamp: off to_stderr: no fileline: off syslog_facility: daemon } amf { mode: disabled } --- # gzip -9 </tmp/config | uuencode config.gz begin 644 config.gz M'XL(`(:%UU("`^U:6V_;-A1^7GX%H>VA`<98-U_DP2^]8D"##N@0#,,`@9)H MAP@EJB3EQ/OU.Y1D6Y+E3G4;U`6,IH8MG<MWSD>1ASS*1$*1<-`_5S\5FG'V M+]%,9*C0/)0D75C.Q+;*7W%>+"P??H`DT5JRJ-!4H2>:A2E;R5(M9#EH3-T; MU[Z93FX<D%::9$FT65@BLZZRTIG[/,Z<7F?^\SCSFLZ6RZVW\?-X\SNAY9*E M3+,U1;E,PU=_OGX9JB!`(E[.[RF1.J)$S\UEXR0G`$"A6"=1*&DLUE1N0B[B MAX4UBI=J9+XJ9Z1(&I&1D;HQ5ZQ*(24965$5/K(L8EFRL#)1(A>Y020U8IFF M<DWXPH*0-$NI*/3"\B;V7@P^^J4<=R>5BHQI(1N"GMTGFE)-(&@F)-.0C*E? M7^_FO,RR8S7RW\W:Z_>W9<)R$M.4/%`YCT6FI>#)@/B"KPYOTI2<]`87S,!- M!_7[N]LPOGO7H1JN-IA>"[Z2>492NK!`UOI6\02#V0KZX['M@W@^N*]>EL'` M?^7.A1M'YY/^V0'<&#*=E'CY.G7G,7PFC<PGA(*KL!?G<T9SD/TN[*4*[VY5 M9]2\99RJC=(T;89`URR&@3."+Z.$J8=1M,$L&24I-B,*W]WB-ZZ%$@93"8#8 MU),(6+?04NE-#KHEEQ82N7D<U<(B,?^U4%2&3V;"^S[CL2<CU<3W+9+R\9Y( MFN`W3@@*NB<]E:L?,T/NM\R0^]D,N3]<AE(F)2A]:89@6A[!7.(<Y@$,_E!) M8'E/Z?'['R1)I-N(OE'0>#?C`*H+EL@PHSHE"@H1=P)7>`$IDV#PGJC[A:5$ M(6/*<EQ]R874%F)+3B(*N,#IH$)D2/S>T(7".P@?K(+&?:BB!-7?Y_0)E#/" M1^;B/@7P*_RRB<3Y;=CCY!R'[\RJ&EQJG%!.-E6PS[%`'5GS@W+\$+FB&D-Q M!;%_!!LY3;J9-`4P479W/OZ+9HT</J5+9FR,J(Y'H#%:IR.C9"%3?YO!N"NI M%U:WI!XV7H8-F/V(*1W04(OALDLIT@'2AV2Z?556?^GK-6K?<1\S,]\8D[1Z MNK#2+'Y@&54PS3CF#N%</.(:\L+2LJ`'1$(>CQ+IGD*D>R'R[(CT3B'2NQ!Y M=D3ZIQ#I7X@\.R+'IQ`YOA!Y=D1.3B%R<B'R[(B<GD+D]$+DV1$Y.X7(V87( MLR,R.(7(X$+DN1'IV"<0:1J/%R*_'Y$Q%QE%,<_VO<E6HW+K?,5%!&8WN,C8 MIP(L+`E78+6,D%.R'NS'=/.V7;VM]4,KW=[:7G_;5VOVV`;9<1RG::?L9^T: M6P.1S%I0JA[3OMLTS$C'1MWP:39_CMOI/YYK6ZN;)9W>R6";!X0UV@N=;L/7 MX:P/Y#OG\Z?;K,^W&T?=1P=O97!H*B#:ZA&-.<-U=Q^W#D);/\"K+#A%O[!D M8355S-66GH4PRY9S]'-ACHH1_61>2P!9E,!SB[A&EFL['K8GV+61X\]M;^[- M_OY_0&[[=&\P(+<'D-\':(J=,7*\N>W,_2&`O/8IU6!`W@D9&D\&`/+;IRV# M`?D]@)S/`H(_9P"@<?O48#"@\0F4C=T!@&;M6GLPH-DI@((!@()VS3@84-`# MR.T'Y`;(#N:V/?=M`!2+/23!JR6QG/A0::ZU9K;GWJXJ%$)F7C<2E>X+=&SZ M<-&QQ\9'QX;+I+V_/49BT*[5KIO+SV<0FXGY'%%7ZT47>+D0E[7%CJ1J;:Y+ MCP/YNW>UP$Y^6U3L=+LZU>I2DK_7JI><YJBX$C*A$L%G=:D6Z1\]U<VN2F>\ M-=;S(XX@Q*J@V>F87&SKG8Y@)_"M9!7T7O3#J[<?PV4#1NG@11^B1G'0+&XZ M2_YUPWA%&.1;]3%6<]&4KTR:$=K.B_%S%B.T`Q9B/@1;)^+L\.Y3OH78_UCL MG[]<BIQ*O:DG8A;A2`BMM"0YKM^&*+<P28S75"KX#9N4&_AGX\DXBF936MZN M7RDP,[4DH%W$NI#FK3A.E&(Q`B<98<#P(]/W*.?%BF77I29]RFD,=1K^5`A9 MI'@M-%7;C5-.P6)>:*RH9%3AE#S5>Z3J)C4Q'+OY2&36?X^;-Q%I!F:UV6*M MH*YO;.5*&<"M,9<IEG0).[5[LZ^;!5,[&#O3:D]'3/V9D0QV<"GL@'=;*KA7 3OY^`=_O%V<2Z^@_U5XFT`"P````` ` end --- # crm_mon -1Arf Last updated: Thu Jan 16 08:11:26 2014 Last change: Wed Jan 15 14:34:52 2014 by root via crm_attribute on o2 Stack: classic openais (with plugin) Current DC: o2 - partition with quorum Version: 1.1.10-65bb87e 4 Nodes configured, 5 expected votes 51 Resources configured Node o1: standby Node o2: standby Node o5: pending OFFLINE: [ o4 ] Full list of resources: Clone Set: cln_O2CB [prm_O2CB] Stopped: [ o1 o2 o4 o5 ] prm_stonith_sbd (stonith:external/sbd): Stopped Clone Set: cln_DLM [prm_DLM] Stopped: [ o1 o2 o4 o5 ] Clone Set: cln_cLVMd [prm_cLVMd] Stopped: [ o1 o2 o4 o5 ] Clone Set: cln_LVM_cVG [prm_LVM_cVG] Stopped: [ o1 o2 o4 o5 ] [...] --- The problem with cLVM mirroring is (besides excessive communication and logging) that there is no persistent "mirror write cache" (bitmap of changed blocks) for a two-device solution (as opposed to MD-RAID (which is not usable for shared devices)). So the few hundred of GB in the volume are resynced all the time. As the disks are iSCSI, that takes at least one hour. During that time simple commands like "lvs" (only one cluster LV) can take as long as two minutes. Without cLVM there is no disk performance issue. Regards, Ulrich > > digimer > > -- > Digimer > Papers and Projects: https://alteeve.ca/w/ > What if the cure for cancer is trapped in the mind of a person without > access to education? > _______________________________________________ > Linux-HA mailing list > Linux-HA@lists.linux-ha.org > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems _______________________________________________ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems