| List: | Commits | « Previous MessageNext Message » | |
| From: | tomas | Date: | April 5 2006 4:05pm |
| Subject: | bk commit into 5.1 tree (tomas:1.2002) BUG#18612 | ||
| View as plain text | |||
Below is the list of changes that have just been committed into a local 5.1 repository of tomas. When tomas does a push these changes will be propagated to the main repository and, within 24 hours after the push, to the public repository. For information on how to access the public repository see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html ChangeSet 1.2002 06/04/05 16:05:42 tomas@stripped +15 -0 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/test/src/NdbRestarts.cpp 1.8 06/04/05 16:05:36 tomas@stripped +1 -2 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/test/run-test/daily-basic-tests.txt 1.35 06/04/05 16:05:36 tomas@stripped +33 -0 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/test/ndbapi/testNodeRestart.cpp 1.23 06/04/05 16:05:36 tomas@stripped +213 -1 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/src/kernel/blocks/qmgr/QmgrMain.cpp 1.28 06/04/05 16:05:36 tomas@stripped +295 -36 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/src/kernel/blocks/qmgr/QmgrInit.cpp 1.9 06/04/05 16:05:36 tomas@stripped +5 -2 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/src/kernel/blocks/qmgr/Qmgr.hpp 1.11 06/04/05 16:05:36 tomas@stripped +25 -2 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp 1.26 06/04/05 16:05:36 tomas@stripped +301 -50 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrInit.cpp 1.11 06/04/05 16:05:36 tomas@stripped +1 -0 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/src/kernel/blocks/ndbcntr/Ndbcntr.hpp 1.8 06/04/05 16:05:36 tomas@stripped +11 -0 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp 1.44 06/04/05 16:05:35 tomas@stripped +28 -0 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/src/kernel/blocks/cmvmi/Cmvmi.cpp 1.26 06/04/05 16:05:35 tomas@stripped +53 -13 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/include/kernel/signaldata/WaitGCP.hpp 1.3 06/04/05 16:05:35 tomas@stripped +5 -2 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/include/kernel/signaldata/StopReq.hpp 1.4 06/04/05 16:05:35 tomas@stripped +22 -16 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/include/kernel/signaldata/FailRep.hpp 1.3 06/04/05 16:05:35 tomas@stripped +13 -3 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced storage/ndb/include/kernel/signaldata/DumpStateOrd.hpp 1.8 06/04/05 16:05:35 tomas@stripped +1 -0 Patches applied to 4.1: ChangeSet@stripped, 2006-04-03 20:43:14+02:00, jonas@stripped ndb - fix testprogam if only 1 node group ChangeSet@stripped, 2006-04-03 13:12:23+02:00, jonas@stripped ndb - Fix compile error...when compiling debug ChangeSet@stripped, 2006-04-03 12:09:50+02:00, jonas@stripped ndb - bug#18612 - partitioned startup add testprg for SR case aswell ChangeSet@stripped, 2006-04-03 11:26:29+02:00, jonas@stripped ndb - bug#18612 post weeked fixes change impl. to use READ_NODESREQ to query state of other qmgr(partition) this as it has no (current) side effects, so that it's possible only to kill starting cluster (if one started and one starting) ChangeSet@stripped, 2006-03-31 18:53:07+02:00, jonas@stripped ndb - autotest add new testpgrom for bug#18612 to autotest ChangeSet@stripped, 2006-03-31 16:46:28+02:00, jonas@stripped ndb - bug#18612 (detection of partitioned cluster) this also impl. gcp safe multi node shutdown 1) block gcp 2) wait for ongoing gcp 3) inform all stopping QMGR's (so that they don't start with error handler) 4) wait for all QMGR's to reply 5) broadcast failrep for stopping nodes 6) (if !master died) unblock gcp ChangeSet@stripped, 2006-03-31 16:36:43+02:00, jonas@stripped ndb - add support for blocking/unblocking GCP using WAIT_GCP_REQ ChangeSet@stripped, 2006-03-31 11:39:35+02:00, jonas@stripped ndb - bug#16447 correct return value in check_cm_cmregreq ChangeSet@stripped, 2006-03-30 14:20:54+02:00, jonas@stripped ndb - bug#15695 bug#16447 bug#18612 For various reasone have a partitioned cluster been created This patch makes sure that when they connect 1) it's detected 2) shutdown is forced # This is a BitKeeper patch. What follows are the unified diffs for the # set of deltas contained in the patch. The rest of the patch, the part # that BitKeeper cares about, is below these diffs. # User: tomas # Host: poseidon.ndb.mysql.com # Root: /home/tomas/wl2325-alcatel --- 1.34/storage/ndb/test/run-test/daily-basic-tests.txt 2005-07-22 15:27:09 +02:00 +++ 1.35/storage/ndb/test/run-test/daily-basic-tests.txt 2006-04-05 16:05:36 +02:00 @@ -413,6 +413,39 @@ cmd: testScan args: -n ScanParallelism +max-time: 500 +cmd: testNodeRestart +args: -n Bug15587 T1 + +max-time: 500 +cmd: testNodeRestart +args: -n Bug15632 T1 + +max-time: 500 +cmd: testNodeRestart +args: -n Bug15685 T1 + +max-time: 500 +cmd: testNodeRestart +args: -n Bug16772 T1 + +#max-time: 500 +#cmd: testSystemRestart +#args: -n Bug18385 T1 +# +max-time: 500 +cmd: testNodeRestart +args: -n Bug18414 T1 + +# OLD FLEX +max-time: 500 +cmd: flexBench +args: -c 25 -t 10 + +max-time: 500 +cmd: flexHammer +args: -r 5 -t 32 + # # DICT TESTS max-time: 1500 --- 1.7/storage/ndb/include/kernel/signaldata/DumpStateOrd.hpp 2006-03-27 18:55:41 +02:00 +++ 1.8/storage/ndb/include/kernel/signaldata/DumpStateOrd.hpp 2006-04-05 16:05:35 +02:00 @@ -64,6 +64,7 @@ // 19 NDBFS Fipple with O_SYNC, O_CREATE etc. // 20-24 BACKUP NdbcntrTestStopOnError = 25, + NdbcntrStopNodes = 70, // 100-105 TUP and ACC // 200-240 UTIL // 300-305 TRIX --- 1.2/storage/ndb/include/kernel/signaldata/FailRep.hpp 2005-04-08 02:43:50 +02:00 +++ 1.3/storage/ndb/include/kernel/signaldata/FailRep.hpp 2006-04-05 16:05:35 +02:00 @@ -18,6 +18,7 @@ #define FAIL_REP_HPP #include "SignalData.hpp" +#include <NodeBitmask.hpp> /** * @@ -27,6 +28,7 @@ * Sender(s) & Reciver(s) */ friend class Qmgr; + friend class Ndbcntr; /** * For printing @@ -35,7 +37,8 @@ public: STATIC_CONST( SignalLength = 2 ); - + STATIC_CONST( ExtraLength = 1 + NdbNodeBitmask::Size ); + enum FailCause { ZOWN_FAILURE=0, ZOTHER_NODE_WHEN_WE_START=1, @@ -43,13 +46,20 @@ ZSTART_IN_REGREQ=3, ZHEARTBEAT_FAILURE=4, ZLINK_FAILURE=5, - ZOTHERNODE_FAILED_DURING_START=6 + ZOTHERNODE_FAILED_DURING_START=6, + ZMULTI_NODE_SHUTDOWN = 7, + ZPARTITIONED_CLUSTER = 8 }; - + private: Uint32 failNodeId; Uint32 failCause; + /** + * Used when failCause == ZPARTITIONED_CLUSTER + */ + Uint32 president; + Uint32 partition[NdbNodeBitmask::Size]; }; --- 1.3/storage/ndb/include/kernel/signaldata/StopReq.hpp 2005-10-06 10:26:06 +02:00 +++ 1.4/storage/ndb/include/kernel/signaldata/StopReq.hpp 2006-04-05 16:05:35 +02:00 @@ -32,7 +32,7 @@ friend class MgmtSrvr; public: - STATIC_CONST( SignalLength = 9 ); + STATIC_CONST( SignalLength = 9 + NdbNodeBitmask::Size); public: Uint32 senderRef; @@ -49,29 +49,34 @@ Int32 readOperationTimeout; // Timeout before read operations are aborted Int32 operationTimeout; // Timeout before all operations are aborted + Uint32 nodes[NdbNodeBitmask::Size]; + static void setSystemStop(Uint32 & requestInfo, bool value); static void setPerformRestart(Uint32 & requestInfo, bool value); static void setNoStart(Uint32 & requestInfo, bool value); static void setInitialStart(Uint32 & requestInfo, bool value); - static void setEscalateOnNodeFail(Uint32 & requestInfo, bool value); /** * Don't perform "graceful" shutdown/restart... */ static void setStopAbort(Uint32 & requestInfo, bool value); + static void setStopNodes(Uint32 & requestInfo, bool value); static bool getSystemStop(const Uint32 & requestInfo); static bool getPerformRestart(const Uint32 & requestInfo); static bool getNoStart(const Uint32 & requestInfo); static bool getInitialStart(const Uint32 & requestInfo); - static bool getEscalateOnNodeFail(const Uint32 & requestInfo); static bool getStopAbort(const Uint32 & requestInfo); + static bool getStopNodes(const Uint32 & requestInfo); }; struct StopConf { STATIC_CONST( SignalLength = 2 ); Uint32 senderData; - Uint32 nodeState; + union { + Uint32 nodeState; + Uint32 nodeId; + }; }; class StopRef @@ -94,7 +99,9 @@ NodeShutdownInProgress = 1, SystemShutdownInProgress = 2, NodeShutdownWouldCauseSystemCrash = 3, - TransactionAbortFailed = 4 + TransactionAbortFailed = 4, + UnsupportedNodeShutdown = 5, + MultiNodeShutdownNotMaster = 6 }; public: @@ -132,16 +139,16 @@ inline bool -StopReq::getEscalateOnNodeFail(const Uint32 & requestInfo) +StopReq::getStopAbort(const Uint32 & requestInfo) { - return requestInfo & 16; + return requestInfo & 32; } inline bool -StopReq::getStopAbort(const Uint32 & requestInfo) +StopReq::getStopNodes(const Uint32 & requestInfo) { - return requestInfo & 32; + return requestInfo & 64; } @@ -187,24 +194,23 @@ inline void -StopReq::setEscalateOnNodeFail(Uint32 & requestInfo, bool value) +StopReq::setStopAbort(Uint32 & requestInfo, bool value) { if(value) - requestInfo |= 16; + requestInfo |= 32; else - requestInfo &= ~16; + requestInfo &= ~32; } inline void -StopReq::setStopAbort(Uint32 & requestInfo, bool value) +StopReq::setStopNodes(Uint32 & requestInfo, bool value) { if(value) - requestInfo |= 32; + requestInfo |= 64; else - requestInfo &= ~32; + requestInfo &= ~64; } - #endif --- 1.2/storage/ndb/include/kernel/signaldata/WaitGCP.hpp 2005-04-08 02:43:52 +02:00 +++ 1.3/storage/ndb/include/kernel/signaldata/WaitGCP.hpp 2006-04-05 16:05:35 +02:00 @@ -46,7 +46,9 @@ Complete = 1, ///< Wait for a GCP to complete CompleteForceStart = 2, ///< Wait for a GCP to complete start one if needed CompleteIfRunning = 3, ///< Wait for ongoing GCP - CurrentGCI = 8 ///< Immediately return current GCI + CurrentGCI = 8, ///< Immediately return current GCI + BlockStartGcp = 9, + UnblockStartGcp = 10 }; Uint32 senderRef; @@ -70,11 +72,12 @@ //friend class Grep::PSCoord; public: - STATIC_CONST( SignalLength = 2 ); + STATIC_CONST( SignalLength = 3 ); public: Uint32 senderData; Uint32 gcp; + Uint32 blockStatus; }; class WaitGCPRef { --- 1.25/storage/ndb/src/kernel/blocks/cmvmi/Cmvmi.cpp 2006-04-05 12:16:10 +02:00 +++ 1.26/storage/ndb/src/kernel/blocks/cmvmi/Cmvmi.cpp 2006-04-05 16:05:35 +02:00 @@ -133,6 +133,9 @@ { } +#ifdef ERROR_INSERT +NodeBitmask c_error_9000_nodes_mask; +#endif void Cmvmi::execNDB_TAMPER(Signal* signal) { @@ -414,21 +417,33 @@ const Uint32 len = signal->getLength(); if(len == 2){ - globalTransporterRegistry.do_connect(tStartingNode); - globalTransporterRegistry.setIOState(tStartingNode, HaltIO); - //----------------------------------------------------- - // Report that the connection to the node is opened - //----------------------------------------------------- - signal->theData[0] = NDB_LE_CommunicationOpened; - signal->theData[1] = tStartingNode; - sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 2, JBB); - //----------------------------------------------------- +#ifdef ERROR_INSERT + if (! (ERROR_INSERTED(9000) && c_error_9000_nodes_mask.get(tStartingNode))) +#endif + { + globalTransporterRegistry.do_connect(tStartingNode); + globalTransporterRegistry.setIOState(tStartingNode, HaltIO); + + //----------------------------------------------------- + // Report that the connection to the node is opened + //----------------------------------------------------- + signal->theData[0] = NDB_LE_CommunicationOpened; + signal->theData[1] = tStartingNode; + sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 2, JBB); + //----------------------------------------------------- + } } else { for(unsigned int i = 1; i < MAX_NODES; i++ ) { jam(); if (i != getOwnNodeId() && getNodeInfo(i).m_type == tData2){ jam(); + +#ifdef ERROR_INSERT + if (ERROR_INSERTED(9000) && c_error_9000_nodes_mask.get(i)) + continue; +#endif + globalTransporterRegistry.do_connect(i); globalTransporterRegistry.setIOState(i, HaltIO); @@ -1034,7 +1049,8 @@ } DumpStateOrd * const & dumpState = (DumpStateOrd *)&signal->theData[0]; - if (dumpState->args[0] == DumpStateOrd::CmvmiDumpConnections){ + Uint32 arg = dumpState->args[0]; + if (arg == DumpStateOrd::CmvmiDumpConnections){ for(unsigned int i = 1; i < MAX_NODES; i++ ){ const char* nodeTypeStr = ""; switch(getNodeInfo(i).m_type){ @@ -1064,13 +1080,13 @@ } } - if (dumpState->args[0] == DumpStateOrd::CmvmiDumpLongSignalMemory){ + if (arg == DumpStateOrd::CmvmiDumpLongSignalMemory){ infoEvent("Cmvmi: g_sectionSegmentPool size: %d free: %d", g_sectionSegmentPool.getSize(), g_sectionSegmentPool.getNoOfFree()); } - if (dumpState->args[0] == DumpStateOrd::CmvmiSetRestartOnErrorInsert) + if (arg == DumpStateOrd::CmvmiSetRestartOnErrorInsert) { if(signal->getLength() == 1) { @@ -1090,7 +1106,7 @@ } } - if (dumpState->args[0] == DumpStateOrd::CmvmiTestLongSigWithDelay) { + if (arg == DumpStateOrd::CmvmiTestLongSigWithDelay) { unsigned i; Uint32 loopCount = dumpState->args[1]; const unsigned len0 = 11; @@ -1117,6 +1133,30 @@ ptr[1].sz = len1; sendSignal(reference(), GSN_TESTSIG, signal, 8, JBB, ptr, 2); } + +#ifdef ERROR_INSERT + if (arg == 9000) + { + SET_ERROR_INSERT_VALUE(9000); + for (Uint32 i = 1; i<signal->getLength(); i++) + c_error_9000_nodes_mask.set(signal->theData[i]); + } + + if (arg == 9001) + { + CLEAR_ERROR_INSERT_VALUE; + for (Uint32 i = 0; i<MAX_NODES; i++) + { + if (c_error_9000_nodes_mask.get(i)) + { + signal->theData[0] = 0; + signal->theData[1] = i; + EXECUTE_DIRECT(CMVMI, GSN_OPEN_COMREQ, signal, 2); + } + } + c_error_9000_nodes_mask.clear(); + } +#endif #ifdef VM_TRACE #if 0 --- 1.43/storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp 2006-03-24 17:11:36 +01:00 +++ 1.44/storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp 2006-04-05 16:05:35 +02:00 @@ -14365,11 +14365,36 @@ jam(); conf->senderData = senderData; conf->gcp = cnewgcp; + conf->blockStatus = cgcpOrderBlocked; sendSignal(senderRef, GSN_WAIT_GCP_CONF, signal, WaitGCPConf::SignalLength, JBB); return; }//if + if (requestType == WaitGCPReq::BlockStartGcp) + { + jam(); + conf->senderData = senderData; + conf->gcp = cnewgcp; + conf->blockStatus = cgcpOrderBlocked; + sendSignal(senderRef, GSN_WAIT_GCP_CONF, signal, + WaitGCPConf::SignalLength, JBB); + cgcpOrderBlocked = 1; + return; + } + + if (requestType == WaitGCPReq::UnblockStartGcp) + { + jam(); + conf->senderData = senderData; + conf->gcp = cnewgcp; + conf->blockStatus = cgcpOrderBlocked; + sendSignal(senderRef, GSN_WAIT_GCP_CONF, signal, + WaitGCPConf::SignalLength, JBB); + cgcpOrderBlocked = 0; + return; + } + if(isMaster()) { /** * Master @@ -14381,6 +14406,7 @@ jam(); conf->senderData = senderData; conf->gcp = coldgcp; + conf->blockStatus = cgcpOrderBlocked; sendSignal(senderRef, GSN_WAIT_GCP_CONF, signal, WaitGCPConf::SignalLength, JBB); return; @@ -14467,6 +14493,7 @@ conf->senderData = ptr.p->clientData; conf->gcp = gcp; + conf->blockStatus = cgcpOrderBlocked; sendSignal(ptr.p->clientRef, GSN_WAIT_GCP_CONF, signal, WaitGCPConf::SignalLength, JBB); @@ -14534,6 +14561,7 @@ c_waitGCPMasterList.next(ptr); conf->senderData = clientData; + conf->blockStatus = cgcpOrderBlocked; sendSignal(clientRef, GSN_WAIT_GCP_CONF, signal, WaitGCPConf::SignalLength, JBB); --- 1.7/storage/ndb/src/kernel/blocks/ndbcntr/Ndbcntr.hpp 2005-10-07 10:58:04 +02:00 +++ 1.8/storage/ndb/src/kernel/blocks/ndbcntr/Ndbcntr.hpp 2006-04-05 16:05:36 +02:00 @@ -204,6 +204,7 @@ void execWAIT_GCP_CONF(Signal* signal); void execSTOP_REQ(Signal* signal); + void execSTOP_CONF(Signal* signal); void execRESUME_REQ(Signal* signal); void execCHANGE_NODE_STATE_CONF(Signal* signal); @@ -339,6 +340,16 @@ void progError(int line, int cause, const char * extra) { cntr.progError(line, cause, extra); } + + enum StopNodesStep { + SR_BLOCK_GCP_START_GCP = 0, + SR_WAIT_COMPLETE_GCP = 1, + SR_UNBLOCK_GCP_START_GCP = 2, + SR_QMGR_STOP_REQ = 3, + SR_WAIT_NODE_FAILURES = 4, + SR_CLUSTER_SHUTDOWN = 12 + } m_state; + SignalCounter m_stop_req_counter; }; private: StopRecord c_stopRec; --- 1.10/storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrInit.cpp 2005-10-07 10:58:04 +02:00 +++ 1.11/storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrInit.cpp 2006-04-05 16:05:36 +02:00 @@ -88,6 +88,7 @@ addRecSignal(GSN_STOP_ME_CONF, &Ndbcntr::execSTOP_ME_CONF); addRecSignal(GSN_STOP_REQ, &Ndbcntr::execSTOP_REQ); + addRecSignal(GSN_STOP_CONF, &Ndbcntr::execSTOP_CONF); addRecSignal(GSN_RESUME_REQ, &Ndbcntr::execRESUME_REQ); addRecSignal(GSN_WAIT_GCP_REF, &Ndbcntr::execWAIT_GCP_REF); --- 1.25/storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp 2006-01-27 17:43:28 +01:00 +++ 1.26/storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp 2006-04-05 16:05:36 +02:00 @@ -42,6 +42,8 @@ #include <signaldata/FsRemoveReq.hpp> #include <signaldata/ReadConfig.hpp> +#include <signaldata/FailRep.hpp> + #include <AttributeHeader.hpp> #include <Configuration.hpp> #include <DebuggerNames.hpp> @@ -1501,13 +1503,74 @@ sendSignal(QMGR_REF, GSN_NODE_FAILREP, signal, NodeFailRep::SignalLength, JBB); + if (c_stopRec.stopReq.senderRef) + { + jam(); + switch(c_stopRec.m_state){ + case StopRecord::SR_WAIT_NODE_FAILURES: + { + jam(); + NdbNodeBitmask tmp; + tmp.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes); + tmp.bitANDC(allFailed); + tmp.copyto(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes); + + if (tmp.isclear()) + { + jam(); + if (c_stopRec.stopReq.senderRef != RNIL) + { + jam(); + StopConf * const stopConf = (StopConf *)&signal->theData[0]; + stopConf->senderData = c_stopRec.stopReq.senderData; + stopConf->nodeState = (Uint32) NodeState::SL_SINGLEUSER; + sendSignal(c_stopRec.stopReq.senderRef, GSN_STOP_CONF, signal, + StopConf::SignalLength, JBB); + } + + c_stopRec.stopReq.senderRef = 0; + WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0]; + req->senderRef = reference(); + req->senderData = StopRecord::SR_UNBLOCK_GCP_START_GCP; + req->requestType = WaitGCPReq::UnblockStartGcp; + sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, + WaitGCPReq::SignalLength, JBA); + } + break; + } + case StopRecord::SR_QMGR_STOP_REQ: + { + NdbNodeBitmask tmp; + tmp.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes); + tmp.bitANDC(allFailed); + + if (tmp.isclear()) + { + Uint32 nodeId = allFailed.find(0); + tmp.set(nodeId); + + StopConf* conf = (StopConf*)signal->getDataPtrSend(); + conf->senderData = c_stopRec.stopReq.senderData; + conf->nodeId = nodeId; + sendSignal(reference(), + GSN_STOP_CONF, signal, StopConf::SignalLength, JBB); + } + + tmp.copyto(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes); + + break; + } + } + } + + signal->theData[0] = NDB_LE_NODE_FAILREP; + signal->theData[2] = 0; + Uint32 nodeId = 0; while(!allFailed.isclear()){ nodeId = allFailed.find(nodeId + 1); allFailed.clear(nodeId); - signal->theData[0] = NDB_LE_NODE_FAILREP; signal->theData[1] = nodeId; - signal->theData[2] = 0; sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 3, JBB); }//for @@ -1951,13 +2014,15 @@ Ndbcntr::execDUMP_STATE_ORD(Signal* signal) { DumpStateOrd * const & dumpState = (DumpStateOrd *)&signal->theData[0]; - if(signal->theData[0] == 13){ + Uint32 arg = dumpState->args[0]; + + if(arg == 13){ infoEvent("Cntr: cstartPhase = %d, cinternalStartphase = %d, block = %d", cstartPhase, cinternalStartphase, cndbBlocksCount); infoEvent("Cntr: cmasterNodeId = %d", cmasterNodeId); } - if (dumpState->args[0] == DumpStateOrd::NdbcntrTestStopOnError){ + if (arg == DumpStateOrd::NdbcntrTestStopOnError){ if (theConfiguration.stopOnError() == true) ((Configuration&)theConfiguration).stopOnError(false); @@ -1970,6 +2035,28 @@ SystemError::SignalLength, JBA); } + if (arg == DumpStateOrd::NdbcntrStopNodes) + { + NdbNodeBitmask mask; + for(Uint32 i = 1; i<signal->getLength(); i++) + mask.set(signal->theData[i]); + + StopReq* req = (StopReq*)signal->getDataPtrSend(); + req->senderRef = RNIL; + req->senderData = 123; + req->requestInfo = 0; + req->singleuser = 0; + req->singleUserApi = 0; + mask.copyto(NdbNodeBitmask::Size, req->nodes); + StopReq::setPerformRestart(req->requestInfo, 1); + StopReq::setNoStart(req->requestInfo, 1); + StopReq::setStopNodes(req->requestInfo, 1); + StopReq::setStopAbort(req->requestInfo, 1); + + sendSignal(reference(), GSN_STOP_REQ, signal, + StopReq::SignalLength, JBB); + return; + } }//Ndbcntr::execDUMP_STATE_ORD() @@ -2030,9 +2117,12 @@ Uint32 senderData = req->senderData; BlockReference senderRef = req->senderRef; bool abort = StopReq::getStopAbort(req->requestInfo); + bool stopnodes = StopReq::getStopNodes(req->requestInfo); - if(getNodeState().startLevel < NodeState::SL_STARTED || - abort && !singleuser){ + if(!singleuser && + (getNodeState().startLevel < NodeState::SL_STARTED || + (abort && !stopnodes))) + { /** * Node is not started yet * @@ -2071,21 +2161,71 @@ else ref->errorCode = StopRef::NodeShutdownInProgress; ref->senderData = senderData; - sendSignal(senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB); + + if (senderRef != RNIL) + sendSignal(senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB); + return; + } + + if (stopnodes && !abort) + { + jam(); + ref->errorCode = StopRef::UnsupportedNodeShutdown; + ref->senderData = senderData; + if (senderRef != RNIL) + sendSignal(senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB); + return; + } + + if (stopnodes && cmasterNodeId != getOwnNodeId()) + { + jam(); + ref->errorCode = StopRef::MultiNodeShutdownNotMaster; + ref->senderData = senderData; + if (senderRef != RNIL) + sendSignal(senderRef, GSN_STOP_REF, signal, StopRef::SignalLength, JBB); return; } c_stopRec.stopReq = * req; c_stopRec.stopInitiatedTime = NdbTick_CurrentMillisecond(); - if(!singleuser) { - if(StopReq::getSystemStop(c_stopRec.stopReq.requestInfo)) { + if (stopnodes) + { + jam(); + + if(!c_stopRec.checkNodeFail(signal)) + { jam(); - if(StopReq::getPerformRestart(c_stopRec.stopReq.requestInfo)){ + return; + } + + char buf[100]; + NdbNodeBitmask mask; + mask.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes); + infoEvent("Initiating shutdown abort of %s", mask.getText(buf)); + ndbout_c("Initiating shutdown abort of %s", mask.getText(buf)); + + WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0]; + req->senderRef = reference(); + req->senderData = StopRecord::SR_BLOCK_GCP_START_GCP; + req->requestType = WaitGCPReq::BlockStartGcp; + sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, + WaitGCPReq::SignalLength, JBB); + return; + } + else if(!singleuser) + { + if(StopReq::getSystemStop(c_stopRec.stopReq.requestInfo)) + { + jam(); + if(StopReq::getPerformRestart(c_stopRec.stopReq.requestInfo)) + { ((Configuration&)theConfiguration).stopOnError(false); } } - if(!c_stopRec.checkNodeFail(signal)){ + if(!c_stopRec.checkNodeFail(signal)) + { jam(); return; } @@ -2155,7 +2295,17 @@ */ NodeBitmask ndbMask; ndbMask.assign(cntr.c_startedNodes); - ndbMask.clear(cntr.getOwnNodeId()); + + if (StopReq::getStopNodes(stopReq.requestInfo)) + { + NdbNodeBitmask tmp; + tmp.assign(NdbNodeBitmask::Size, stopReq.nodes); + ndbMask.bitANDC(tmp); + } + else + { + ndbMask.clear(cntr.getOwnNodeId()); + } CheckNodeGroups* sd = (CheckNodeGroups*)&signal->theData[0]; sd->blockRef = cntr.reference(); @@ -2177,7 +2327,8 @@ ref->errorCode = StopRef::NodeShutdownWouldCauseSystemCrash; const BlockReference bref = stopReq.senderRef; - cntr.sendSignal(bref, GSN_STOP_REF, signal, StopRef::SignalLength, JBB); + if (bref != RNIL) + cntr.sendSignal(bref, GSN_STOP_REF, signal, StopRef::SignalLength, JBB); stopReq.senderRef = 0; @@ -2227,23 +2378,23 @@ if(stopReq.getSystemStop(stopReq.requestInfo) || stopReq.singleuser){ jam(); if(stopReq.singleuser) - { - jam(); - AbortAllReq * req = (AbortAllReq*)&signal->theData[0]; - req->senderRef = cntr.reference(); - req->senderData = 12; - cntr.sendSignal(DBTC_REF, GSN_ABORT_ALL_REQ, signal, - AbortAllReq::SignalLength, JBB); - } + { + jam(); + AbortAllReq * req = (AbortAllReq*)&signal->theData[0]; + req->senderRef = cntr.reference(); + req->senderData = 12; + cntr.sendSignal(DBTC_REF, GSN_ABORT_ALL_REQ, signal, + AbortAllReq::SignalLength, JBB); + } else - { - WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0]; - req->senderRef = cntr.reference(); - req->senderData = 12; - req->requestType = WaitGCPReq::CompleteForceStart; - cntr.sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, - WaitGCPReq::SignalLength, JBB); - } + { + WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0]; + req->senderRef = cntr.reference(); + req->senderData = StopRecord::SR_CLUSTER_SHUTDOWN; + req->requestType = WaitGCPReq::CompleteForceStart; + cntr.sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, + WaitGCPReq::SignalLength, JBB); + } } else { jam(); StopPermReq * req = (StopPermReq*)&signal->theData[0]; @@ -2405,7 +2556,7 @@ WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0]; req->senderRef = reference(); - req->senderData = 12; + req->senderData = StopRecord::SR_CLUSTER_SHUTDOWN; req->requestType = WaitGCPReq::CompleteForceStart; sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, WaitGCPReq::SignalLength, JBB); @@ -2414,29 +2565,129 @@ void Ndbcntr::execWAIT_GCP_CONF(Signal* signal){ jamEntry(); - ndbrequire(StopReq::getSystemStop(c_stopRec.stopReq.requestInfo)); - NodeState newState(NodeState::SL_STOPPING_3, true); + WaitGCPConf* conf = (WaitGCPConf*)signal->getDataPtr(); - /** - * Inform QMGR so that arbitrator won't kill us - */ - NodeStateRep * rep = (NodeStateRep *)&signal->theData[0]; - rep->nodeState = newState; - rep->nodeState.masterNodeId = cmasterNodeId; - rep->nodeState.setNodeGroup(c_nodeGroup); - EXECUTE_DIRECT(QMGR, GSN_NODE_STATE_REP, signal, NodeStateRep::SignalLength); - - if(StopReq::getPerformRestart(c_stopRec.stopReq.requestInfo)){ - jam(); - StartOrd * startOrd = (StartOrd *)&signal->theData[0]; - startOrd->restartInfo = c_stopRec.stopReq.requestInfo; - sendSignalWithDelay(CMVMI_REF, GSN_START_ORD, signal, 500, - StartOrd::SignalLength); - } else { + switch(conf->senderData){ + case StopRecord::SR_BLOCK_GCP_START_GCP: + { + jam(); + /** + * + */ + if(!c_stopRec.checkNodeFail(signal)) + { + jam(); + goto unblock; + } + + WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0]; + req->senderRef = reference(); + req->senderData = StopRecord::SR_WAIT_COMPLETE_GCP; + req->requestType = WaitGCPReq::CompleteIfRunning; + + sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, + WaitGCPReq::SignalLength, JBB); + return; + } + case StopRecord::SR_UNBLOCK_GCP_START_GCP: + { jam(); - sendSignalWithDelay(CMVMI_REF, GSN_STOP_ORD, signal, 500, 1); + return; + } + case StopRecord::SR_WAIT_COMPLETE_GCP: + { + jam(); + if(!c_stopRec.checkNodeFail(signal)) + { + jam(); + goto unblock; + } + + NdbNodeBitmask tmp; + tmp.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes); + c_stopRec.m_stop_req_counter = tmp; + NodeReceiverGroup rg(QMGR, tmp); + StopReq * stopReq = (StopReq *)&signal->theData[0]; + * stopReq = c_stopRec.stopReq; + stopReq->senderRef = reference(); + sendSignal(rg, GSN_STOP_REQ, signal, StopReq::SignalLength, JBA); + c_stopRec.m_state = StopRecord::SR_QMGR_STOP_REQ; + return; + } + case StopRecord::SR_CLUSTER_SHUTDOWN: + { + jam(); + break; + } + } + + { + ndbrequire(StopReq::getSystemStop(c_stopRec.stopReq.requestInfo)); + NodeState newState(NodeState::SL_STOPPING_3, true); + + /** + * Inform QMGR so that arbitrator won't kill us + */ + NodeStateRep * rep = (NodeStateRep *)&signal->theData[0]; + rep->nodeState = newState; + rep->nodeState.masterNodeId = cmasterNodeId; + rep->nodeState.setNodeGroup(c_nodeGroup); + EXECUTE_DIRECT(QMGR, GSN_NODE_STATE_REP, signal, + NodeStateRep::SignalLength); + + if(StopReq::getPerformRestart(c_stopRec.stopReq.requestInfo)){ + jam(); + StartOrd * startOrd = (StartOrd *)&signal->theData[0]; + startOrd->restartInfo = c_stopRec.stopReq.requestInfo; + sendSignalWithDelay(CMVMI_REF, GSN_START_ORD, signal, 500, + StartOrd::SignalLength); + } else { + jam(); + sendSignalWithDelay(CMVMI_REF, GSN_STOP_ORD, signal, 500, 1); + } + return; + } + +unblock: + WaitGCPReq * req = (WaitGCPReq*)&signal->theData[0]; + req->senderRef = reference(); + req->senderData = StopRecord::SR_UNBLOCK_GCP_START_GCP; + req->requestType = WaitGCPReq::UnblockStartGcp; + sendSignal(DBDIH_REF, GSN_WAIT_GCP_REQ, signal, + WaitGCPReq::SignalLength, JBB); +} + +void +Ndbcntr::execSTOP_CONF(Signal* signal) +{ + jamEntry(); + StopConf *conf = (StopConf*)signal->getDataPtr(); + ndbrequire(c_stopRec.m_state == StopRecord::SR_QMGR_STOP_REQ); + c_stopRec.m_stop_req_counter.clearWaitingFor(conf->nodeId); + if (c_stopRec.m_stop_req_counter.done()) + { + char buf[100]; + NdbNodeBitmask mask; + mask.assign(NdbNodeBitmask::Size, c_stopRec.stopReq.nodes); + infoEvent("Stopping of %s", mask.getText(buf)); + ndbout_c("Stopping of %s", mask.getText(buf)); + + /** + * Kill any node... + */ + FailRep * const failRep = (FailRep *)&signal->theData[0]; + failRep->failCause = FailRep::ZMULTI_NODE_SHUTDOWN; + NodeReceiverGroup rg(QMGR, c_clusterNodes); + Uint32 nodeId = 0; + while ((nodeId = NdbNodeBitmask::find(c_stopRec.stopReq.nodes, nodeId+1)) + != NdbNodeBitmask::NotFound) + { + failRep->failNodeId = nodeId; + sendSignal(rg, GSN_FAIL_REP, signal, FailRep::SignalLength, JBA); + } + c_stopRec.m_state = StopRecord::SR_WAIT_NODE_FAILURES; + return; } - return; } void Ndbcntr::execSTTORRY(Signal* signal){ --- 1.10/storage/ndb/src/kernel/blocks/qmgr/Qmgr.hpp 2006-01-27 17:43:28 +01:00 +++ 1.11/storage/ndb/src/kernel/blocks/qmgr/Qmgr.hpp 2006-04-05 16:05:36 +02:00 @@ -29,6 +29,7 @@ #include <signaldata/CmRegSignalData.hpp> #include <signaldata/ApiRegSignalData.hpp> #include <signaldata/FailRep.hpp> +#include <signaldata/StopReq.hpp> #include <signaldata/AllocNodeId.hpp> #include <SafeCounter.hpp> @@ -104,7 +105,12 @@ }; struct StartRecord { - void reset(){ m_startKey++; m_startNode = 0;} + void reset(){ + m_startKey++; + m_startNode = 0; + m_gsn = RNIL; + m_nodes.clearWaitingFor(); + } Uint32 m_startKey; Uint32 m_startNode; Uint64 m_startTimeout; @@ -116,6 +122,14 @@ NdbNodeBitmask c_definedNodes; // DB nodes in config NdbNodeBitmask c_clusterNodes; // DB nodes in cluster NodeBitmask c_connectedNodes; // All kinds of connected nodes + + /** + * Nodes which we're checking for partitioned cluster + * + * i.e. nodes that connect to use, when we already have elected president + */ + NdbNodeBitmask c_readnodes_nodes; + Uint32 c_maxDynamicId; // Records @@ -208,6 +222,7 @@ void execPRES_TOCONF(Signal* signal); void execDISCONNECT_REP(Signal* signal); void execSYSTEM_ERROR(Signal* signal); + void execSTOP_REQ(Signal* signal); // Received signals void execDUMP_STATE_ORD(Signal* signal); @@ -222,6 +237,8 @@ void execREAD_NODESREQ(Signal* signal); void execSET_VAR_REQ(Signal* signal); + void execREAD_NODESREF(Signal* signal); + void execREAD_NODESCONF(Signal* signal); void execAPI_VERSION_REQ(Signal* signal); void execAPI_BROADCAST_REP(Signal* signal); @@ -244,6 +261,8 @@ void execARBIT_STOPREP(Signal* signal); // Statement blocks + void check_readnodes_reply(Signal* signal, Uint32 nodeId, Uint32 gsn); + void node_failed(Signal* signal, Uint16 aFailedNode); void checkStartInterface(Signal* signal); void failReport(Signal* signal, @@ -261,8 +280,9 @@ // Generated statement blocks void startphase1(Signal* signal); - void electionWon(); + void electionWon(Signal* signal); void cmInfoconf010Lab(Signal* signal); + void apiHbHandlingLab(Signal* signal); void timerHandlingLab(Signal* signal); void hbReceivedLab(Signal* signal); @@ -397,6 +417,9 @@ Uint16 cfailedNodes[MAX_NDB_NODES]; Uint16 cprepFailedNodes[MAX_NDB_NODES]; Uint16 ccommitFailedNodes[MAX_NDB_NODES]; + + StopReq c_stopReq; + void check_multi_node_shutdown(Signal* signal); struct OpAllocNodeIdReq { RequestTracker m_tracker; --- 1.8/storage/ndb/src/kernel/blocks/qmgr/QmgrInit.cpp 2006-01-27 17:43:28 +01:00 +++ 1.9/storage/ndb/src/kernel/blocks/qmgr/QmgrInit.cpp 2006-04-05 16:05:36 +02:00 @@ -35,9 +35,8 @@ Uint32 hbDBAPI = 500; setHbApiDelay(hbDBAPI); - - c_connectedNodes.clear(); c_connectedNodes.set(getOwnNodeId()); + c_stopReq.senderRef = 0; }//Qmgr::initData() void Qmgr::initRecords() @@ -52,6 +51,7 @@ // Transit signals addRecSignal(GSN_DUMP_STATE_ORD, &Qmgr::execDUMP_STATE_ORD); + addRecSignal(GSN_STOP_REQ, &Qmgr::execSTOP_REQ); addRecSignal(GSN_DEBUG_SIG, &Qmgr::execDEBUG_SIG); addRecSignal(GSN_CONTINUEB, &Qmgr::execCONTINUEB); addRecSignal(GSN_CM_HEARTBEAT, &Qmgr::execCM_HEARTBEAT); @@ -101,6 +101,9 @@ addRecSignal(GSN_ARBIT_CHOOSEREF, &Qmgr::execARBIT_CHOOSEREF); addRecSignal(GSN_ARBIT_STOPREP, &Qmgr::execARBIT_STOPREP); + addRecSignal(GSN_READ_NODESREF, &Qmgr::execREAD_NODESREF); + addRecSignal(GSN_READ_NODESCONF, &Qmgr::execREAD_NODESCONF); + initData(); }//Qmgr::Qmgr() --- 1.27/storage/ndb/src/kernel/blocks/qmgr/QmgrMain.cpp 2006-03-24 17:11:37 +01:00 +++ 1.28/storage/ndb/src/kernel/blocks/qmgr/QmgrMain.cpp 2006-04-05 16:05:36 +02:00 @@ -56,6 +56,33 @@ #define DEBUG_START3(signal, msg) #endif +/** + * c_start.m_gsn = GSN_CM_REGREQ + * Possible for all nodes + * c_start.m_nodes contains all nodes in config + * + * c_start.m_gsn = GSN_CM_NODEINFOREQ; + * Set when receiving CM_REGCONF + * State possible for starting node only (not in cluster) + * + * c_start.m_nodes contains all node in alive cluster that + * that has not replied to GSN_CM_NODEINFOREQ + * passed by president in GSN_CM_REGCONF + * + * c_start.m_gsn = GSN_CM_ADD + * Possible for president only + * Set when receiving and accepting CM_REGREQ (to include node) + * + * c_start.m_nodes contains all nodes in alive cluster + starting node + * that has not replied to GSN_CM_ADD + * by sending GSN_CM_ACKADD + * + * c_start.m_gsn = GSN_CM_NODEINFOCONF + * Possible for non presidents only + * c_start.m_nodes contains a node that has been accepted by president + * but has not connected to us yet + */ + // Signal entries and statement blocks /* 4 P R O G R A M */ /*******************************/ @@ -280,18 +307,24 @@ { jamEntry(); const Uint32 nodeId = signal->theData[0]; + + if (ERROR_INSERTED(931)) + { + jam(); + ndbout_c("Discarding CONNECT_REP(%d)", nodeId); + infoEvent("Discarding CONNECT_REP(%d)", nodeId); + return; + } + c_connectedNodes.set(nodeId); NodeRecPtr nodePtr; nodePtr.i = getOwnNodeId(); ptrCheckGuard(nodePtr, MAX_NODES, nodeRec); switch(nodePtr.p->phase){ - case ZSTARTING: case ZRUNNING: + ndbrequire(!c_clusterNodes.get(nodeId)); + case ZSTARTING: jam(); - if(!c_start.m_nodes.isWaitingFor(nodeId)){ - jam(); - return; - } break; case ZPREPARE_FAIL: case ZFAIL_CLOSING: @@ -303,35 +336,83 @@ case ZAPI_INACTIVE: return; } - + + if (getNodeInfo(nodeId).getType() != NodeInfo::DB) + { + jam(); + return; + } + switch(c_start.m_gsn){ case GSN_CM_REGREQ: jam(); sendCmRegReq(signal, nodeId); + + /** + * We're waiting for CM_REGCONF c_start.m_nodes contains all configured + * nodes + */ + ndbrequire(nodePtr.p->phase == ZSTARTING); + ndbrequire(c_start.m_nodes.isWaitingFor(nodeId)); return; case GSN_CM_NODEINFOREQ: jam(); - sendCmNodeInfoReq(signal, nodeId, nodePtr.p); + + if (c_start.m_nodes.isWaitingFor(nodeId)) + { + jam(); + ndbrequire(getOwnNodeId() != cpresident); + ndbrequire(nodePtr.p->phase == ZSTARTING); + sendCmNodeInfoReq(signal, nodeId, nodePtr.p); + return; + } return; - case GSN_CM_ADD:{ + case GSN_CM_NODEINFOCONF:{ jam(); - - ndbrequire(getOwnNodeId() != cpresident); - c_start.m_nodes.clearWaitingFor(nodeId); - c_start.m_gsn = RNIL; - NodeRecPtr addNodePtr; - addNodePtr.i = nodeId; - ptrCheckGuard(addNodePtr, MAX_NDB_NODES, nodeRec); - cmAddPrepare(signal, addNodePtr, nodePtr.p); - return; + ndbrequire(getOwnNodeId() != cpresident); + ndbrequire(nodePtr.p->phase == ZRUNNING); + if (c_start.m_nodes.isWaitingFor(nodeId)) + { + jam(); + c_start.m_nodes.clearWaitingFor(nodeId); + c_start.m_gsn = RNIL; + + NodeRecPtr addNodePtr; + addNodePtr.i = nodeId; + ptrCheckGuard(addNodePtr, MAX_NDB_NODES, nodeRec); + cmAddPrepare(signal, addNodePtr, nodePtr.p); + return; + } } default: - return; + (void)1; } + + ndbrequire(!c_start.m_nodes.isWaitingFor(nodeId)); + ndbrequire(!c_readnodes_nodes.get(nodeId)); + c_readnodes_nodes.set(nodeId); + signal->theData[0] = reference(); + sendSignal(calcQmgrBlockRef(nodeId), GSN_READ_NODESREQ, signal, 1, JBA); return; }//Qmgr::execCONNECT_REP() +void +Qmgr::execREAD_NODESCONF(Signal* signal) +{ + check_readnodes_reply(signal, + refToNode(signal->getSendersBlockRef()), + GSN_READ_NODESCONF); +} + +void +Qmgr::execREAD_NODESREF(Signal* signal) +{ + check_readnodes_reply(signal, + refToNode(signal->getSendersBlockRef()), + GSN_READ_NODESREF); +} + /*******************************/ /* CM_INFOCONF */ /*******************************/ @@ -622,22 +703,33 @@ jamEntry(); const CmRegConf * const cmRegConf = (CmRegConf *)&signal->theData[0]; + Uint32 presidentNodeId = cmRegConf->presidentNodeId; if (!ndbCompatible_ndb_ndb(NDB_VERSION, cmRegConf->presidentVersion)) { jam(); char buf[128]; - BaseString::snprintf(buf,sizeof(buf),"incompatible version own=0x%x other=0x%x, shutting down", NDB_VERSION, cmRegConf->presidentVersion); + BaseString::snprintf(buf,sizeof(buf), + "incompatible version own=0x%x other=0x%x, " + " shutting down", + NDB_VERSION, cmRegConf->presidentVersion); systemErrorLab(signal, __LINE__, buf); return; } - + myNodePtr.i = getOwnNodeId(); + ptrCheckGuard(myNodePtr, MAX_NDB_NODES, nodeRec); + + ndbrequire(c_start.m_gsn == GSN_CM_REGREQ); + ndbrequire(myNodePtr.p->phase = ZSTARTING); + cpdistref = cmRegConf->presidentBlockRef; cpresident = cmRegConf->presidentNodeId; UintR TdynamicId = cmRegConf->dynamicId; c_maxDynamicId = TdynamicId; c_clusterNodes.assign(NdbNodeBitmask::Size, cmRegConf->allNdbNodes); + myNodePtr.p->ndynamicId = TdynamicId; + /*--------------------------------------------------------------*/ // Send this as an EVENT REPORT to inform about hearing about // other NDB node proclaiming to be president. @@ -648,10 +740,6 @@ signal->theData[3] = TdynamicId; sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 4, JBB); - myNodePtr.i = getOwnNodeId(); - ptrCheckGuard(myNodePtr, MAX_NDB_NODES, nodeRec); - myNodePtr.p->ndynamicId = TdynamicId; - for (nodePtr.i = 1; nodePtr.i < MAX_NDB_NODES; nodePtr.i++) { jam(); if (c_clusterNodes.get(nodePtr.i)){ @@ -675,6 +763,84 @@ }//Qmgr::execCM_REGCONF() void +Qmgr::check_readnodes_reply(Signal* signal, Uint32 nodeId, Uint32 gsn) +{ + NodeRecPtr myNodePtr; + myNodePtr.i = getOwnNodeId(); + ptrCheckGuard(myNodePtr, MAX_NDB_NODES, nodeRec); + + NodeRecPtr nodePtr; + nodePtr.i = nodeId; + ptrCheckGuard(nodePtr, MAX_NDB_NODES, nodeRec); + + ndbrequire(c_readnodes_nodes.get(nodeId)); + ReadNodesConf* conf = (ReadNodesConf*)signal->getDataPtr(); + if (gsn == GSN_READ_NODESREF) + { + jam(); +retry: + signal->theData[0] = reference(); + sendSignal(calcQmgrBlockRef(nodeId), GSN_READ_NODESREQ, signal, 1, JBA); + return; + } + + if (conf->masterNodeId == ZNIL) + { + jam(); + goto retry; + } + + Uint32 president = conf->masterNodeId; + if (president == cpresident) + { + jam(); + c_readnodes_nodes.clear(nodeId); + return; + } + + char buf[255]; + BaseString::snprintf(buf, sizeof(buf), + "Partitioned cluster! check StartPartialTimeout, " + " node %d thinks %d is president, " + " I think president is: %d", + nodeId, president, cpresident); + + ndbout_c(buf); + CRASH_INSERTION(933); + + if (getNodeState().startLevel == NodeState::SL_STARTED) + { + jam(); + NdbNodeBitmask part; + part.assign(NdbNodeBitmask::Size, conf->clusterNodes); + FailRep* rep = (FailRep*)signal->getDataPtrSend(); + rep->failCause = FailRep::ZPARTITIONED_CLUSTER; + rep->president = cpresident; + c_clusterNodes.copyto(NdbNodeBitmask::Size, rep->partition); + Uint32 ref = calcQmgrBlockRef(nodeId); + Uint32 i = 0; + while((i = part.find(i + 1)) != NdbNodeBitmask::NotFound) + { + if (i == nodeId) + continue; + rep->failNodeId = i; + sendSignal(ref, GSN_FAIL_REP, signal, FailRep::SignalLength, JBA); + } + rep->failNodeId = nodeId; + sendSignal(ref, GSN_FAIL_REP, signal, FailRep::SignalLength, JBB); + return; + } + + CRASH_INSERTION(932); + + progError(__LINE__, + NDBD_EXIT_ARBIT_SHUTDOWN, + buf); + + ndbrequire(false); +} + +void Qmgr::sendCmNodeInfoReq(Signal* signal, Uint32 nodeId, const NodeRec * self){ CmNodeInfoReq * const req = (CmNodeInfoReq*)signal->getDataPtrSend(); req->nodeId = getOwnNodeId(); @@ -706,13 +872,15 @@ void Qmgr::execCM_REGREF(Signal* signal) { jamEntry(); - c_regReqReqRecv++; - // Ignore block reference in data[0] UintR TaddNodeno = signal->theData[1]; UintR TrefuseReason = signal->theData[2]; Uint32 candidate = signal->theData[3]; DEBUG_START3(signal, TrefuseReason); + + c_regReqReqRecv++; + + // Ignore block reference in data[0] if(candidate != cpresidentCandidate){ jam(); @@ -800,7 +968,7 @@ Uint64 now = NdbTick_CurrentMillisecond(); if((c_regReqReqRecv == cnoOfNodes) || now > c_stopElectionTime){ jam(); - electionWon(); + electionWon(signal); sendSttorryLab(signal); /** @@ -814,7 +982,7 @@ }//Qmgr::execCM_REGREF() void -Qmgr::electionWon(){ +Qmgr::electionWon(Signal* signal){ NodeRecPtr myNodePtr; cpresident = getOwnNodeId(); /* This node becomes president. */ myNodePtr.i = getOwnNodeId(); @@ -833,6 +1001,12 @@ cpresidentAlive = ZTRUE; c_stopElectionTime = ~0; c_start.reset(); + + signal->theData[0] = NDB_LE_CM_REGCONF; + signal->theData[1] = getOwnNodeId(); + signal->theData[2] = cpresident; + signal->theData[3] = 1; + sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 4, JBB); } /* @@ -967,7 +1141,7 @@ ndbrequire(signal->header.theVerId_signalNumber == GSN_CM_ADD); c_start.m_nodes.clearWaitingFor(); c_start.m_nodes.setWaitingFor(nodePtr.i); - c_start.m_gsn = GSN_CM_ADD; + c_start.m_gsn = GSN_CM_NODEINFOCONF; #else warningEvent("Enabling communication to CM_ADD node %u state=%d", nodePtr.i, @@ -1872,7 +2046,8 @@ const Uint32 nodeId = rep->nodeId; const Uint32 err = rep->err; c_connectedNodes.clear(nodeId); - + c_readnodes_nodes.clear(nodeId); + NodeRecPtr nodePtr; nodePtr.i = getOwnNodeId(); ptrCheckGuard(nodePtr, MAX_NODES, nodeRec); @@ -1893,9 +2068,13 @@ case ZAPI_ACTIVE: ndbrequire(false); case ZAPI_INACTIVE: + { + char buf[100]; + BaseString::snprintf(buf, 100, "Node %u disconected", nodeId); + progError(__LINE__, NDBD_EXIT_SR_OTHERNODEFAILED, buf); ndbrequire(false); } - + } node_failed(signal, nodeId); }//DISCONNECT_REP @@ -2152,10 +2331,16 @@ failedNodePtr.i = aFailedNode; ptrCheckGuard(failedNodePtr, MAX_NODES, nodeRec); + FailRep* rep = (FailRep*)signal->getDataPtr(); + + check_multi_node_shutdown(signal); + if (failedNodePtr.i == getOwnNodeId()) { jam(); + Uint32 code = 0; const char * msg = 0; + char extra[100]; switch(aFailCause){ case FailRep::ZOWN_FAILURE: msg = "Own failure"; @@ -2176,17 +2361,51 @@ case FailRep::ZLINK_FAILURE: msg = "Connection failure"; break; + case FailRep::ZPARTITIONED_CLUSTER: + { + code = NDBD_EXIT_ARBIT_SHUTDOWN; + char buf1[100], buf2[100]; + c_clusterNodes.getText(buf1); + if (signal->getLength()== FailRep::SignalLength + FailRep::ExtraLength && + signal->header.theVerId_signalNumber == GSN_FAIL_REP) + { + jam(); + NdbNodeBitmask part; + part.assign(NdbNodeBitmask::Size, rep->partition); + part.getText(buf2); + BaseString::snprintf(extra, sizeof(extra), + "Partitioned cluster!" + " Our cluster: %s other cluster: %s", + buf1, buf2); + } + else + { + jam(); + BaseString::snprintf(extra, sizeof(extra), + "Partitioned cluster!" + " Our cluster: %s ", buf1); + } + msg = extra; + break; + } + case FailRep::ZMULTI_NODE_SHUTDOWN: + msg = "Multi node shutdown"; + break; + default: + msg = "<UNKNOWN>"; } - char buf[100]; - BaseString::snprintf(buf, 100, + CRASH_INSERTION(932); + + char buf[255]; + BaseString::snprintf(buf, sizeof(buf), "We(%u) have been declared dead by %u reason: %s(%u)", getOwnNodeId(), refToNode(signal->getSendersBlockRef()), aFailCause, msg ? msg : "<Unknown>"); - - progError(__LINE__, 0, buf); + + progError(__LINE__, code, buf); return; }//if @@ -2243,7 +2462,9 @@ { NodeRecPtr myNodePtr; jamEntry(); - + + check_multi_node_shutdown(signal); + PrepFailReqRef * const prepFail = (PrepFailReqRef *)&signal->theData[0]; BlockReference Tblockref = prepFail->xxxBlockRef; @@ -3895,6 +4116,7 @@ if (! (arbitRec.getTimediff() > getArbitTimeout())) return; #endif + CRASH_INSERTION(932); progError(__LINE__, NDBD_EXIT_ARBIT_SHUTDOWN, "Arbitrator decided to shutdown this node"); } @@ -4055,6 +4277,43 @@ NodeReceiverGroup rg(API_CLUSTERMGR, mask); sendSignal(rg, api.gsn, signal, len, JBB); // forward sections +} + +void +Qmgr::execSTOP_REQ(Signal* signal) +{ + jamEntry(); + c_stopReq = * (StopReq*)signal->getDataPtr(); + + if (c_stopReq.senderRef) + { + ndbrequire(NdbNodeBitmask::get(c_stopReq.nodes, getOwnNodeId())); + + StopConf *conf = (StopConf*)signal->getDataPtrSend(); + conf->senderData = c_stopReq.senderData; + conf->nodeState = getOwnNodeId(); + sendSignal(c_stopReq.senderRef, + GSN_STOP_CONF, signal, StopConf::SignalLength, JBA); + } +} + +void +Qmgr::check_multi_node_shutdown(Signal* signal) +{ + if (c_stopReq.senderRef && + NdbNodeBitmask::get(c_stopReq.nodes, getOwnNodeId())) + { + jam(); + if(StopReq::getPerformRestart(c_stopReq.requestInfo)) + { + jam(); + StartOrd * startOrd = (StartOrd *)&signal->theData[0]; + startOrd->restartInfo = c_stopReq.requestInfo; + EXECUTE_DIRECT(CMVMI, GSN_START_ORD, signal, 2); + } else { + EXECUTE_DIRECT(CMVMI, GSN_STOP_ORD, signal, 1); + } + } } void --- 1.22/storage/ndb/test/ndbapi/testNodeRestart.cpp 2006-04-05 12:16:10 +02:00 +++ 1.23/storage/ndb/test/ndbapi/testNodeRestart.cpp 2006-04-05 16:05:36 +02:00 @@ -22,7 +22,7 @@ #include <NdbRestarts.hpp> #include <Vector.hpp> #include <signaldata/DumpStateOrd.hpp> - +#include <Bitmask.hpp> int runLoadTable(NDBT_Context* ctx, NDBT_Step* step){ @@ -669,6 +669,206 @@ return NDBT_FAILED; } +int +runBug18612(NDBT_Context* ctx, NDBT_Step* step){ + + // Assume two replicas + NdbRestarter restarter; + if (restarter.getNumDbNodes() < 2) + { + ctx->stopTest(); + return NDBT_OK; + } + + Uint32 cnt = restarter.getNumDbNodes(); + + for(int loop = 0; loop < ctx->getNumLoops(); loop++) + { + int partition0[256]; + int partition1[256]; + bzero(partition0, sizeof(partition0)); + bzero(partition1, sizeof(partition1)); + Bitmask<4> nodesmask; + + Uint32 node1 = restarter.getDbNodeId(rand()%cnt); + for (Uint32 i = 0; i<cnt/2; i++) + { + do { + int tmp = restarter.getRandomNodeOtherNodeGroup(node1, rand()); + if (tmp == -1) + break; + node1 = tmp; + } while(nodesmask.get(node1)); + + partition0[i] = node1; + partition1[i] = restarter.getRandomNodeSameNodeGroup(node1, rand()); + + ndbout_c("nodes %d %d", node1, partition1[i]); + + assert(!nodesmask.get(node1)); + assert(!nodesmask.get(partition1[i])); + nodesmask.set(node1); + nodesmask.set(partition1[i]); + } + + ndbout_c("done"); + + int dump[255]; + dump[0] = DumpStateOrd::NdbcntrStopNodes; + memcpy(dump + 1, partition0, sizeof(int)*cnt/2); + + Uint32 master = restarter.getMasterNodeId(); + + if (restarter.dumpStateOneNode(master, dump, 1+cnt/2)) + return NDBT_FAILED; + + if (restarter.waitNodesNoStart(partition0, cnt/2)) + return NDBT_FAILED; + + int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 }; + + if (restarter.dumpStateAllNodes(val2, 2)) + return NDBT_FAILED; + + if (restarter.insertErrorInAllNodes(932)) + return NDBT_FAILED; + + dump[0] = 9000; + memcpy(dump + 1, partition0, sizeof(int)*cnt/2); + for (Uint32 i = 0; i<cnt/2; i++) + if (restarter.dumpStateOneNode(partition1[i], dump, 1+cnt/2)) + return NDBT_FAILED; + + dump[0] = 9000; + memcpy(dump + 1, partition1, sizeof(int)*cnt/2); + for (Uint32 i = 0; i<cnt/2; i++) + if (restarter.dumpStateOneNode(partition0[i], dump, 1+cnt/2)) + return NDBT_FAILED; + + if (restarter.startNodes(partition0, cnt/2)) + return NDBT_FAILED; + + if (restarter.waitNodesStartPhase(partition0, cnt/2, 2)) + return NDBT_FAILED; + + dump[0] = 9001; + for (Uint32 i = 0; i<cnt/2; i++) + if (restarter.dumpStateAllNodes(dump, 2)) + return NDBT_FAILED; + + if (restarter.waitNodesNoStart(partition0, cnt/2)) + return NDBT_FAILED; + + for (Uint32 i = 0; i<cnt/2; i++) + if (restarter.restartOneDbNode(partition0[i], true, true, true)) + return NDBT_FAILED; + + if (restarter.waitNodesNoStart(partition0, cnt/2)) + return NDBT_FAILED; + + if (restarter.startAll()) + return NDBT_FAILED; + + if (restarter.waitClusterStarted()) + return NDBT_FAILED; + } + return NDBT_OK; +} + +int +runBug18612SR(NDBT_Context* ctx, NDBT_Step* step){ + + // Assume two replicas + NdbRestarter restarter; + if (restarter.getNumDbNodes() < 2) + { + ctx->stopTest(); + return NDBT_OK; + } + + Uint32 cnt = restarter.getNumDbNodes(); + + for(int loop = 0; loop < ctx->getNumLoops(); loop++) + { + int partition0[256]; + int partition1[256]; + bzero(partition0, sizeof(partition0)); + bzero(partition1, sizeof(partition1)); + Bitmask<4> nodesmask; + + Uint32 node1 = restarter.getDbNodeId(rand()%cnt); + for (Uint32 i = 0; i<cnt/2; i++) + { + do { + int tmp = restarter.getRandomNodeOtherNodeGroup(node1, rand()); + if (tmp == -1) + break; + node1 = tmp; + } while(nodesmask.get(node1)); + + partition0[i] = node1; + partition1[i] = restarter.getRandomNodeSameNodeGroup(node1, rand()); + + ndbout_c("nodes %d %d", node1, partition1[i]); + + assert(!nodesmask.get(node1)); + assert(!nodesmask.get(partition1[i])); + nodesmask.set(node1); + nodesmask.set(partition1[i]); + } + + ndbout_c("done"); + + if (restarter.restartAll(false, true, false)) + return NDBT_FAILED; + + int dump[255]; + dump[0] = 9000; + memcpy(dump + 1, partition0, sizeof(int)*cnt/2); + for (Uint32 i = 0; i<cnt/2; i++) + if (restarter.dumpStateOneNode(partition1[i], dump, 1+cnt/2)) + return NDBT_FAILED; + + dump[0] = 9000; + memcpy(dump + 1, partition1, sizeof(int)*cnt/2); + for (Uint32 i = 0; i<cnt/2; i++) + if (restarter.dumpStateOneNode(partition0[i], dump, 1+cnt/2)) + return NDBT_FAILED; + + int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 }; + + if (restarter.dumpStateAllNodes(val2, 2)) + return NDBT_FAILED; + + if (restarter.insertErrorInAllNodes(932)) + return NDBT_FAILED; + + if (restarter.startAll()) + return NDBT_FAILED; + + if (restarter.waitClusterStartPhase(2)) + return NDBT_FAILED; + + dump[0] = 9001; + for (Uint32 i = 0; i<cnt/2; i++) + if (restarter.dumpStateAllNodes(dump, 2)) + return NDBT_FAILED; + + if (restarter.waitClusterNoStart(30)) + if (restarter.waitNodesNoStart(partition0, cnt/2, 10)) + if (restarter.waitNodesNoStart(partition1, cnt/2, 10)) + return NDBT_FAILED; + + if (restarter.startAll()) + return NDBT_FAILED; + + if (restarter.waitClusterStarted()) + return NDBT_FAILED; + } + return NDBT_OK; +} + + NDBT_TESTSUITE(testNodeRestart); TESTCASE("NoLoad", "Test that one node at a time can be stopped and then restarted "\ @@ -961,6 +1161,18 @@ "Test bug with NF during NR"){ INITIALIZER(runLoadTable); STEP(runBug18414); + FINALIZER(runClearTable); +} +TESTCASE("Bug18612", + "Test bug with partitioned clusters"){ + INITIALIZER(runLoadTable); + STEP(runBug18612); + FINALIZER(runClearTable); +} +TESTCASE("Bug18612SR", + "Test bug with partitioned clusters"){ + INITIALIZER(runLoadTable); + STEP(runBug18612SR); FINALIZER(runClearTable); } NDBT_TESTSUITE_END(testNodeRestart); --- 1.7/storage/ndb/test/src/NdbRestarts.cpp 2006-04-05 12:16:11 +02:00 +++ 1.8/storage/ndb/test/src/NdbRestarts.cpp 2006-04-05 16:05:36 +02:00 @@ -445,8 +445,7 @@ << ") secs " << endl; NdbSleep_SecSleep(seconds); - randomId = (rand() % _restarter.getNumDbNodes()); - nodeId = _restarter.getDbNodeId(randomId); + nodeId = _restarter.getRandomNodeOtherNodeGroup(nodeId, rand()); g_info << _restart->m_name << ": node = "<< nodeId << endl; CHECK(_restarter.insertErrorInNode(nodeId, 9999) == 0,
| Thread | ||
|---|---|---|
| • bk commit into 5.1 tree (tomas:1.2002) BUG#18612 | tomas | 5 Apr |
© 1995-2008 MySQL AB, 2008-2009 Sun Microsystems, Inc.
Page generated in 1.063 sec. using MySQL 5.1.14-beta-log