Greenplum failed segment的恢复方法

【前记】

Segment检测及故障切换机制
GP Master首先会检测Primary状态，如果Primary不可连通，那么将会检测Mirror状态，Primary/Mirror状态总共有4种：
1. Primary活着，Mirror活着。GP Master探测Primary成功之后直接返回，进行下一个Segment检测；
2. Primary活着，Mirror挂了。GP Master探测Primary成功之后，通过Primary返回的状态得知Mirror挂掉了（Mirror挂掉之后，Primary将会探测到，将自己变成ChangeTracking模式），这时候更新Master元信息，进行下一个Segment检测；
3. Primary挂了，Mirror活着。GP Master探测Primary失败之后探测Mirror，发现Mirror是活着，这时候更新Master上面的元信息，同时使Mirror接管Primary（故障切换），进行下一个Segment检测；
4. Primary挂了，Mirror挂了。GP Master探测Primary失败之后探测Mirror，Mirror也是挂了，直到重试最大值，结束这个Segment的探测，也不更新Master元信息了，进行下一个Segment检测。
上面的2-4需要进行gprecoverseg进行segment恢复。

对失败的segment节点；启动时会直接跳过，忽略。

[gpadmin@mdw ~]$ gpstart
20160718:18:43:27:002949 gpstart:mdw:gpadmin-[INFO]:-Starting gpstart with args:
20160718:18:43:27:002949 gpstart:mdw:gpadmin-[INFO]:-Gathering information and validating the environment...
20160718:18:43:27:002949 gpstart:mdw:gpadmin-[INFO]:-Greenplum Binary Version: 'postgres (Greenplum Database) 4.3.8.1 build 1'
20160718:18:43:28:002949 gpstart:mdw:gpadmin-[INFO]:-Greenplum Catalog Version: '201310150'
20160718:18:43:28:002949 gpstart:mdw:gpadmin-[INFO]:-Starting Master instance in admin mode
20160718:18:43:30:002949 gpstart:mdw:gpadmin-[INFO]:-Obtaining Greenplum Master catalog information
20160718:18:43:30:002949 gpstart:mdw:gpadmin-[INFO]:-Obtaining Segment details from master...
20160718:18:43:30:002949 gpstart:mdw:gpadmin-[INFO]:-Setting new master era
20160718:18:43:30:002949 gpstart:mdw:gpadmin-[INFO]:-Master Started...
20160718:18:43:30:002949 gpstart:mdw:gpadmin-[INFO]:-Shutting down master
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[WARNING]:-Skipping startup of segment marked down in configuration: on sdw2 directory /home/gpadmin/gpdata/gpdatam/gpseg0 <<<<<
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:---------------------------
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-Master instance parameters
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:---------------------------
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-Database                 = template1
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-Master Port              = 1921
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-Master directory         = /home/gpadmin/gpdata/pgmaster/gpseg-1
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-Timeout                  = 600 seconds
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-Master standby           = Off
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:---------------------------------------
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-Segment instances that will be started
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:---------------------------------------
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-   Host   Datadir                               Port    Role
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-   sdw1   /home/gpadmin/gpdata/gpdatap/gpseg0   40000   Primary
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-   sdw2   /home/gpadmin/gpdata/gpdatap/gpseg1   40000   Primary
20160718:18:43:32:002949 gpstart:mdw:gpadmin-[INFO]:-   sdw1   /home/gpadmin/gpdata/gpdatam/gpseg1   50000   Mirror

Continue with Greenplum instance startup Yy|Nn (default=N):
> y
20160718:18:43:34:002949 gpstart:mdw:gpadmin-[INFO]:-Commencing parallel primary and mirror segment instance startup, please wait...
...........
20160718:18:43:45:002949 gpstart:mdw:gpadmin-[INFO]:-Process results...
20160718:18:43:45:002949 gpstart:mdw:gpadmin-[INFO]:-----------------------------------------------------
20160718:18:43:45:002949 gpstart:mdw:gpadmin-[INFO]:-   Successful segment starts                                            = 3
20160718:18:43:45:002949 gpstart:mdw:gpadmin-[INFO]:-   Failed segment starts                                                = 0
20160718:18:43:45:002949 gpstart:mdw:gpadmin-[WARNING]:-Skipped segment starts (segments are marked down in configuration)   = 1   <<<<<<<<
20160718:18:43:45:002949 gpstart:mdw:gpadmin-[INFO]:-----------------------------------------------------
20160718:18:43:45:002949 gpstart:mdw:gpadmin-[INFO]:-
20160718:18:43:46:002949 gpstart:mdw:gpadmin-[INFO]:-Successfully started 3 of 3 segment instances, skipped 1 other segments
20160718:18:43:46:002949 gpstart:mdw:gpadmin-[INFO]:-----------------------------------------------------
20160718:18:43:46:002949 gpstart:mdw:gpadmin-[WARNING]:-****************************************************************************
20160718:18:43:46:002949 gpstart:mdw:gpadmin-[WARNING]:-There are 1 segment(s) marked down in the database
20160718:18:43:46:002949 gpstart:mdw:gpadmin-[WARNING]:-To recover from this current state, review usage of the gprecoverseg
20160718:18:43:46:002949 gpstart:mdw:gpadmin-[WARNING]:-management utility which will recover failed segment instance databases.
20160718:18:43:46:002949 gpstart:mdw:gpadmin-[WARNING]:-****************************************************************************
20160718:18:43:46:002949 gpstart:mdw:gpadmin-[INFO]:-Starting Master instance mdw directory /home/gpadmin/gpdata/pgmaster/gpseg-1
20160718:18:43:48:002949 gpstart:mdw:gpadmin-[INFO]:-Command pg_ctl reports Master mdw instance active
20160718:18:43:49:002949 gpstart:mdw:gpadmin-[INFO]:-No standby master configured.  skipping...
20160718:18:43:49:002949 gpstart:mdw:gpadmin-[WARNING]:-Number of segments not attempted to start: 1
20160718:18:43:49:002949 gpstart:mdw:gpadmin-[INFO]:-Check status of database with gpstate utility

查看数据库的mirror的节点启动状态

[gpadmin@mdw ~]$ gpstate -m
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:-Starting gpstate with args: -m
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:-local Greenplum Version: 'postgres (Greenplum Database) 4.3.8.1 build 1'
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:-master Greenplum Version: 'PostgreSQL 8.2.15 (Greenplum Database 4.3.8.1 build 1) on x86_64-unknown-linux-gnu, compiled by GCC gcc (GCC) 4.4.2 compiled on Apr 20 2016 08:08:56'
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:-Obtaining Segment details from master...
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:--------------------------------------------------------------
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:--Current GPDB mirror list and status
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:--Type = Spread
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:--------------------------------------------------------------
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:-   Mirror   Datadir                               Port    Status    Data Status
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[WARNING]:-sdw2     /home/gpadmin/gpdata/gpdatam/gpseg0   50000   Failed                   <<<<<<<<
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:-   sdw1     /home/gpadmin/gpdata/gpdatam/gpseg1   50000   Passive   Synchronized
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[INFO]:--------------------------------------------------------------
20160718:18:45:48:003084 gpstate:mdw:gpadmin-[WARNING]:-1 segment(s) configured as mirror(s) have failed

可直观看出“[WARNING]:-sdw2 /home/gpadmin/gpdata/gpdatam/gpseg0 50000 Failed ”

如何恢复这个mirror segment呢？当然primary segment也是这样恢复的

1. 首先产生一个恢复的配置文件 ： gprecoverseg -o ./recov

[gpadmin@mdw ~]$ gprecoverseg -o ./recov
20160718:18:47:04:003134 gprecoverseg:mdw:gpadmin-[INFO]:-Starting gprecoverseg with args: -o ./recov
20160718:18:47:04:003134 gprecoverseg:mdw:gpadmin-[INFO]:-local Greenplum Version: 'postgres (Greenplum Database) 4.3.8.1 build 1'
20160718:18:47:04:003134 gprecoverseg:mdw:gpadmin-[INFO]:-master Greenplum Version: 'PostgreSQL 8.2.15 (Greenplum Database 4.3.8.1 build 1) on x86_64-unknown-linux-gnu, compiled by GCC gcc (GCC) 4.4.2 compiled on Apr 20 2016 08:08:56'
20160718:18:47:04:003134 gprecoverseg:mdw:gpadmin-[INFO]:-Checking if segments are ready
20160718:18:47:04:003134 gprecoverseg:mdw:gpadmin-[INFO]:-Obtaining Segment details from master...
20160718:18:47:06:003134 gprecoverseg:mdw:gpadmin-[INFO]:-Obtaining Segment details from master...
20160718:18:47:07:003134 gprecoverseg:mdw:gpadmin-[INFO]:-Configuration file output to ./recov successfully.

2. 查看恢复的配置文件；可以知道哪些segment需要恢复

[gpadmin@mdw ~]$ cat recov
filespaceOrder=fastdisk
sdw2:50000:/home/gpadmin/gpdata/gpdatam/gpseg0

3. 使用这个配置文件进行恢复 ： gprecoverseg -i ./recov

[gpadmin@mdw ~]$ gprecoverseg -i ./recov
20160718:18:47:33:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Starting gprecoverseg with args: -i ./recov
20160718:18:47:34:003187 gprecoverseg:mdw:gpadmin-[INFO]:-local Greenplum Version: 'postgres (Greenplum Database) 4.3.8.1 build 1'
20160718:18:47:34:003187 gprecoverseg:mdw:gpadmin-[INFO]:-master Greenplum Version: 'PostgreSQL 8.2.15 (Greenplum Database 4.3.8.1 build 1) on x86_64-unknown-linux-gnu, compiled by GCC gcc (GCC) 4.4.2 compiled on Apr 20 2016 08:08:56'
20160718:18:47:34:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Checking if segments are ready
20160718:18:47:34:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Obtaining Segment details from master...
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Obtaining Segment details from master...
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Greenplum instance recovery parameters
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:----------------------------------------------------------
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Recovery from configuration -i option supplied
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:----------------------------------------------------------
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Recovery 1 of 1
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:----------------------------------------------------------
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Synchronization mode                          = Incremental
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Failed instance host                          = sdw2
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Failed instance address                       = sdw2
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Failed instance directory                     = /home/gpadmin/gpdata/gpdatam/gpseg0
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Failed instance port                          = 50000
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Failed instance replication port              = 51000
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Failed instance fastdisk directory            = /data/gpdata/seg1/pg_mir_cdr/gpseg0
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Recovery Source instance host                 = sdw1
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Recovery Source instance address              = sdw1
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Recovery Source instance directory            = /home/gpadmin/gpdata/gpdatap/gpseg0
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Recovery Source instance port                 = 40000
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Recovery Source instance replication port     = 41000
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Recovery Source instance fastdisk directory   = /data/gpdata/seg1/pg_pri_cdr/gpseg0
20160718:18:47:35:003187 gprecoverseg:mdw:gpadmin-[INFO]:-   Recovery Target                               = in-place
20160718:18:48:06:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Process results...
20160718:18:48:06:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Done updating primaries
20160718:18:48:06:003187 gprecoverseg:mdw:gpadmin-[INFO]:-******************************************************************
20160718:18:48:06:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Updating segments for resynchronization is completed.
20160718:18:48:06:003187 gprecoverseg:mdw:gpadmin-[INFO]:-For segments updated successfully, resynchronization will continue in the background.
20160718:18:48:06:003187 gprecoverseg:mdw:gpadmin-[INFO]:-
20160718:18:48:06:003187 gprecoverseg:mdw:gpadmin-[INFO]:-Use  gpstate -s  to check the resynchronization progress.
20160718:18:48:06:003187 gprecoverseg:mdw:gpadmin-[INFO]:-******************************************************************

4. 查看恢复状态

[gpadmin@mdw ~]$ gpstate -m
20160718:18:48:39:003353 gpstate:mdw:gpadmin-[INFO]:-Starting gpstate with args: -m
20160718:18:48:39:003353 gpstate:mdw:gpadmin-[INFO]:-local Greenplum Version: 'postgres (Greenplum Database) 4.3.8.1 build 1'
20160718:18:48:39:003353 gpstate:mdw:gpadmin-[INFO]:-master Greenplum Version: 'PostgreSQL 8.2.15 (Greenplum Database 4.3.8.1 build 1) on x86_64-unknown-linux-gnu, compiled by GCC gcc (GCC) 4.4.2 compiled on Apr 20 2016 08:08:56'
20160718:18:48:39:003353 gpstate:mdw:gpadmin-[INFO]:-Obtaining Segment details from master...
20160718:18:48:40:003353 gpstate:mdw:gpadmin-[INFO]:--------------------------------------------------------------
20160718:18:48:40:003353 gpstate:mdw:gpadmin-[INFO]:--Current GPDB mirror list and status
20160718:18:48:40:003353 gpstate:mdw:gpadmin-[INFO]:--Type = Spread
20160718:18:48:40:003353 gpstate:mdw:gpadmin-[INFO]:--------------------------------------------------------------
20160718:18:48:40:003353 gpstate:mdw:gpadmin-[INFO]:-   Mirror   Datadir                               Port    Status    Data Status
20160718:18:48:40:003353 gpstate:mdw:gpadmin-[INFO]:-   sdw2     /home/gpadmin/gpdata/gpdatam/gpseg0   50000   Passive   Resynchronizing
20160718:18:48:40:003353 gpstate:mdw:gpadmin-[INFO]:-   sdw1     /home/gpadmin/gpdata/gpdatam/gpseg1   50000   Passive   Synchronized
20160718:18:48:40:003353 gpstate:mdw:gpadmin-[INFO]:--------------------------------------------------------------

5. 到上一步，数据库的主备就恢复了，但是还有一步，是可选的。
你要不要把primary , mirror角色对调一下，因为现在mirror和primary和优先角色是相反的。
如果要对调，使用以下命令，会停库来处理。

gprecoverseg -r

【总结】

用于修复Segment的是gprecoverseg。使用方式比较简单，有限的几个主要参数如下：
-i ：主要参数，用于指定一个配置文件，该配置文件描述了需要修复的Segment和修复后的目的位置。
-F ：可选项，指定后，gprecoverseg会将”-i”中指定的或标记”d”的实例删除，并从活着的Mirror复制一个完整一份到目标位置。
-r ：当FTS发现有Primary宕机并进行主备切换，在gprecoverseg修复后，担当Primary的Mirror角色并不会立即切换回来，就会导致部分主机上活跃的Segment过多从而引起性能瓶颈。因此需要恢复Segment原先的角色，称为re-balance。