There is a work in progress script to automate a safe master failover from on server to another.
https://gerrit.wikimedia.org/r/#/c/operations/software/wmfmariadbpy/+/439871/
Do some production testing on passive hosts (codfw) before using for real on es1014 and db1052.
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Invalid | ayounsi | T199142 Increase network capacity (2018-19 Q1 Goal) | |||
| Resolved | • Cmjohnson | T183585 Rack/cable/configure asw2-b-eqiad switch stack | |||
| Resolved | • jcrespo | T197073 switchover es1014 to es1017 | |||
| Resolved | None | T186320 Decommission db1051-db1060 (DBA tracking) | |||
| Resolved | • Marostegui | T197069 Failover db1052 (s1) db primary master | |||
| Resolved | LSobanski | T156461 [META ticket] Automation for our DBs tracking task | |||
| Resolved | • jcrespo | T199224 Test database master switchover script on codfw |
Some progress reporting:
$ python3 Python 3.4.2 (default, Oct 8 2014, 10:45:20) [GCC 4.9.1] on linux Type "help", "copyright", "credits" or "license" for more information. >>> import WMFMariaDB >>> master=WMFMariaDB.WMFMariaDB(host='db1095') >>> import WMFReplication >>> master_replication=WMFReplication.WMFReplication(master) >>> master_replication.debug() db1095.eqiad.wmnet:3306/(none)> master: db1067.eqiad.wmnet:3306, io: Yes, sql: Yes, pos: db1067-bin.001280/704908984 >>> master_replication.stop_slave() {'master_ssl_cipher': '', 'master_server_id': 171974720, 'master_ssl_key': '', 'last_errno': 0, 'last_sql_errno': 0, 'replicate_ignore_server_ids': '', 'last_io_error': '', 'replicate_wild_ignore_table': '', 'replicate_ignore_domain_ids': '', 'relay_master_log_file': 'db1067-bin.001280', 'slave_io_running': 'No', 'replicate_wild_do_table': '', 'master_ssl_verify_server_cert': 'No', 'gtid_io_pos': '', 'seconds_behind_master': None, 'master_host': 'db1067.eqiad.wmnet', 'parallel_mode': 'conservative', 'relay_log_pos': 99104951, 'relay_log_space': 100145534, 'using_gtid': 'No', 'skip_counter': 0, 'read_master_log_pos': 705765165, 'success': True, 'master_port': 3306, 'relay_log_file': 'db1095-relay-bin.000018', 'exec_master_log_pos': 705765165, 'until_log_file': '', 'master_ssl_allowed': 'Yes', 'master_user': 'repl', 'master_ssl_cert': '', 'until_log_pos': 0, 'replicate_do_domain_ids': '', 'replicate_ignore_table': '', 'last_sql_error': '', 'replicate_ignore_db': '', 'master_ssl_ca_file': '', 'replicate_do_table': '', 'connect_retry': 60, 'until_condition': 'None', 'slave_sql_running': 'No', 'last_io_errno': 0, 'master_log_file': 'db1067-bin.001280', 'last_error': '', 'replicate_do_db': '', 'master_ssl_crlpath': '', 'master_ssl_crl': '', 'slave_io_state': '', 'master_ssl_ca_path': ''} >>> master_replication.debug() db1095.eqiad.wmnet:3306/(none)> master: db1067.eqiad.wmnet:3306, io: No, sql: No, pos: db1067-bin.001280/705765165 >>> master_replication.debug() db1095.eqiad.wmnet:3306/(none)> master: db1067.eqiad.wmnet:3306, io: No, sql: No, pos: db1067-bin.001280/705765165 >>> master_replication.reset_slave() {'database': None, 'success': True, 'numrows': 0, 'port': 3306, 'query': 'RESET SLAVE ALL', 'rows': None, 'host': 'db1095.eqiad.wmnet', 'fields': None} >>> master_replication.debug() db1095.eqiad.wmnet:3306/(none)> Not configured as a slave
0.7 seconds on performing a switchover with no writes ongoing (easy case):
root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ time ./switchover.py db1095 db1102 Starting preflight checks... Setting up original master as read-only Slave caught up to the master after waiting 0.008965492248535156 seconds Servers sync at master: db1095-bin.000006:576902809 slave: db1102-bin.000001:576902795 Stopping original master->slave replication Setting up replica as read-write All commands where successful, current status: original master read_only: 1 / original slave read_only: 0 Trying to invert replication direction Verifying everything went as expected... SUCCESS: Master switch completed successfully real 0m0.710s user 0m0.220s sys 0m0.008s root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ time ./switchover.py db1102 db1095 Starting preflight checks... Setting up original master as read-only Slave caught up to the master after waiting 0.0075452327728271484 seconds Servers sync at master: db1102-bin.000001:576902795 slave: db1095-bin.000006:576902809 Stopping original master->slave replication Setting up replica as read-write All commands where successful, current status: original master read_only: 1 / original slave read_only: 0 Trying to invert replication direction Verifying everything went as expected... SUCCESS: Master switch completed successfully real 0m0.697s user 0m0.208s sys 0m0.016s
Change 447365 had a related patch set uploaded (by Jcrespo; owner: Jcrespo):
[operations/puppet@production] mariadb: Migrate db1118 from mysql 8 to mariadb 10.1
Change 447365 merged by Jcrespo:
[operations/puppet@production] mariadb: Migrate db1118 from mysql 8 to mariadb 10.1
What the first tests on production-like hosts look like:
root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ time ./move_replica.py db1118 db1095 Purging binary logs to speed up coordinate finding and waiting 5.0 seconds... Stopping replication on the instance Stopped at master position: db1102-bin.000015:30633013, slave position db1118-bin.000007:82334 Last GTID executed: BEGIN GTID 171974720-171974720-31485244 Position found on the master at: db1095-bin.000021:82334 SUCCESS: Topology change completed successfully real 0m5.605s user 0m0.308s sys 0m0.028s root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ time ./move_replica.py db1118 db1102 Purging binary logs to speed up coordinate finding and waiting 5.0 seconds... Stopping replication on the instance Stopped at master position: db1095-bin.000021:731731, slave position db1118-bin.000008:253914 Last GTID executed: BEGIN GTID 171974720-171974720-31486754 Position found on the master at: db1102-bin.000016:253930 SUCCESS: Topology change completed successfully real 0m5.717s user 0m0.416s sys 0m0.060s root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ time ./move_replica.py db1118 db1095 Purging binary logs to speed up coordinate finding and waiting 5.0 seconds... Stopping replication on the instance Stopped at master position: db1102-bin.000016:1024067, slave position db1118-bin.000009:452398 Last GTID executed: BEGIN GTID 171974720-171974720-31488149 Position found on the master at: db1095-bin.000022:452398 SUCCESS: Topology change completed successfully real 0m5.717s user 0m0.420s sys 0m0.020s root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ time ./move_replica.py db1118 db1102 Purging binary logs to speed up coordinate finding and waiting 5.0 seconds... Stopping replication on the instance Stopped at master position: db1095-bin.000022:1519934, slave position db1118-bin.000010:140801 Last GTID executed: BEGIN GTID 171974720-171974720-31489628 Position found on the master at: db1102-bin.000014:11087634 SUCCESS: Topology change completed successfully real 1m30.972s user 1m17.256s sys 0m3.680s
This last one was incorrectly detected, so checking now that.
root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ for host in db1095 db1102 db1118; do mysql.py -h $host -e "set sql_log_bin=0; FLUSH BINAR Y LOGS;"; done; time ./switchover.py db1095 db1118 Starting preflight checks... * Original read only values are as expected (master: read_only=0, slave: read_only=1) * The host to fail over is a direct replica of the master * Replication is up and running between the 2 hosts * The replication lag is acceptable: 0 (lower than the configured or default timeout) * The master is not a replica of any other host Testing if to migrate db1118.eqiad.wmnet:3306/(none)... Nope Testing if to migrate db1102.eqiad.wmnet:3306/(none)... Stopped at master position: db1095-bin.000046:384, slave position db1102-bin.000037:400 Last GTID not executed: BEGIN GTID 171974784-171974784-30197 Position found on the master at: db1118-bin.000034:400 Migrated db1102.eqiad.wmnet:3306/(none) successfully from master to replica Setting up original master as read-only Slave caught up to the master after waiting 0.010107278823852539 seconds Servers sync at master: db1095-bin.000046:1124 slave: db1118-bin.000034:998 Stopping original master->slave replication Setting up replica as read-write All commands where successful, current status: original master read_only: 1 / original slave read_only: 0 Trying to invert replication direction Verifying everything went as expected... SUCCESS: Master switch completed successfully real 0m1.820s user 0m0.368s sys 0m0.040s root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ for host in db1095 db1102 db1118; do mysql.py -h $host -e "set sql_log_bin=0; FLUSH BINARY LOGS;"; done; time ./switchover.py db1118 db1095 Starting preflight checks... * Original read only values are as expected (master: read_only=0, slave: read_only=1) * The host to fail over is a direct replica of the master * Replication is up and running between the 2 hosts * The replication lag is acceptable: 0 (lower than the configured or default timeout) * The master is not a replica of any other host Testing if to migrate db1102.eqiad.wmnet:3306/(none)... Stopped at master position: db1118-bin.000035:786, slave position db1102-bin.000038:715 Last GTID not executed: BEGIN GTID 171970572-171970572-35 Position found on the master at: db1095-bin.000047:699 Migrated db1102.eqiad.wmnet:3306/(none) successfully from master to replica Testing if to migrate db1095.eqiad.wmnet:3306/(none)... Nope Setting up original master as read-only Slave caught up to the master after waiting 0.0103302001953125 seconds Servers sync at master: db1118-bin.000035:1156 slave: db1095-bin.000047:998 Stopping original master->slave replication Setting up replica as read-write All commands where successful, current status: original master read_only: 1 / original slave read_only: 0 Trying to invert replication direction Verifying everything went as expected... SUCCESS: Master switch completed successfully real 0m1.794s user 0m0.364s sys 0m0.028s
Change 447595 had a related patch set uploaded (by Jcrespo; owner: Jcrespo):
[operations/software/wmfmariadbpy@master] switchover: Make posible replica migration an optional, separate step
Change 447595 merged by Jcrespo:
[operations/software/wmfmariadbpy@master] switchover: Make posible replica migration an optional, separate step
Change 447603 had a related patch set uploaded (by Jcrespo; owner: Jcrespo):
[operations/software/wmfmariadbpy@master] switchover: Add the functionality to start and stop heartbeat
It works now fully, including pt-heartbeat handling:
root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ ./switchover.py --skip-slave-move db1102 db1095 Starting preflight checks... * Original read only values are as expected (master: read_only=0, slave: read_only=1) * The host to fail over is a direct replica of the master * Replication is up and running between the 2 hosts * The replication lag is acceptable: 0 (lower than the configured or default timeout) * The master is not a replica of any other host ----- OUTPUT of '/bin/ps --no-hea...pid,args -C perl' ----- 9405 /usr/bin/perl /usr/local/bin/pt-heartbeat-wikimedia --defaults-file=/root/.my.cnf -D heartbeat --shard=test-s1 --datacenter=eqiad --update --replace --interval=1 --set-vars=binlog_format=STATEMENT -S /run/mysqld/mysqld.sock --daemonize --pid /var/run/pt-heartbeat.pid ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 3.20hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/bin/ps --no-hea...pid,args -C perl'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. Stopping heartbeat pid 9405 at db1102.eqiad.wmnet:3306/(none) ----- OUTPUT of '/bin/kill 9405' ----- ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 3.58hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/bin/kill 9405'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. Setting up original master as read-only Slave caught up to the master after waiting 0.010645627975463867 seconds Servers sync at master: db1102-bin.000040:1011408046 slave: db1095-bin.000049:1011453017 Stopping original master->slave replication Setting up replica as read-write All commands where successful, current status: original master read_only: 1 / original slave read_only: 0 Trying to invert replication direction Starting heartbeat section test-s1 at db1095.eqiad.wmnet ----- OUTPUT of '/usr/bin/nohup /...d &> /dev/null &' ----- ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 3.98hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/usr/bin/nohup /...d &> /dev/null &'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. ----- OUTPUT of '/bin/ps --no-hea...pid,args -C perl' ----- 17695 /usr/bin/perl /usr/local/bin/pt-heartbeat-wikimedia --defaults-file=/root/.my.cnf -D heartbeat --shard=test-s1 --datacenter=eqiad --update --replace --interval=1 --set-vars=binlog_format=STATEMENT -S /run/mysqld/mysqld.sock --daemonize --pid /var/run/pt-heartbeat.pid ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 3.90hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/bin/ps --no-hea...pid,args -C perl'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. Detected heartbeat at db1095.eqiad.wmnet running with PID 17695 Verifying everything went as expected... SUCCESS: Master switch completed successfully root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ ./switchover.py --skip-slave-move db1095 db1102 Starting preflight checks... * Original read only values are as expected (master: read_only=0, slave: read_only=1) * The host to fail over is a direct replica of the master * Replication is up and running between the 2 hosts * The replication lag is acceptable: 0 (lower than the configured or default timeout) * The master is not a replica of any other host ----- OUTPUT of '/bin/ps --no-hea...pid,args -C perl' ----- 17695 /usr/bin/perl /usr/local/bin/pt-heartbeat-wikimedia --defaults-file=/root/.my.cnf -D heartbeat --shard=test-s1 --datacenter=eqiad --update --replace --interval=1 --set-vars=binlog_format=STATEMENT -S /run/mysqld/mysqld.sock --daemonize --pid /var/run/pt-heartbeat.pid ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 3.66hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/bin/ps --no-hea...pid,args -C perl'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. Stopping heartbeat pid 17695 at db1095.eqiad.wmnet:3306/(none) ----- OUTPUT of '/bin/kill 17695' ----- ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 4.10hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/bin/kill 17695'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. Setting up original master as read-only Slave caught up to the master after waiting 0.01145792007446289 seconds Servers sync at master: db1095-bin.000049:1011467915 slave: db1102-bin.000040:1011420097 Stopping original master->slave replication Setting up replica as read-write All commands where successful, current status: original master read_only: 1 / original slave read_only: 0 Trying to invert replication direction Starting heartbeat section test-s1 at db1102.eqiad.wmnet ----- OUTPUT of '/usr/bin/nohup /...d &> /dev/null &' ----- ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 3.43hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/usr/bin/nohup /...d &> /dev/null &'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. ----- OUTPUT of '/bin/ps --no-hea...pid,args -C perl' ----- 9747 /usr/bin/perl /usr/local/bin/pt-heartbeat-wikimedia --defaults-file=/root/.my.cnf -D heartbeat --shard=test-s1 --datacenter=eqiad --update --replace --interval=1 --set-vars=binlog_format=STATEMENT -S /run/mysqld/mysqld.sock --daemonize --pid /var/run/pt-heartbeat.pid ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 3.55hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/bin/ps --no-hea...pid,args -C perl'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. Detected heartbeat at db1102.eqiad.wmnet running with PID 9747 Verifying everything went as expected... SUCCESS: Master switch completed successfully
This is brilliant. A great step towards getting the failover automation we have always talked about.
Excellent job!
Looking forward to seeing this working tomorrow.
Change 447603 merged by Jcrespo:
[operations/software/wmfmariadbpy@master] switchover: Add the functionality to start and stop heartbeat
Codfw testing:
root@neodymium:~/wmfmariadbpy/wmfmariadbpy$ ./switchover.py --skip-slave-move es2018 es2017 Starting preflight checks... * Original read only values are as expected (master: read_only=0, slave: read_only=1) * The host to fail over is a direct replica of the master * Replication is up and running between the 2 hosts * The replication lag is acceptable: 0 (lower than the configured or default timeout) * The master is not a replica of any other host ----- OUTPUT of '/bin/ps --no-hea...pid,args -C perl' ----- ================ PASS: | | 0% (0/1) [00:01<?, ?hosts/s] FAIL: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:01<00:00, 1.10s/hosts] 100.0% (1/1) of nodes failed to execute command '/bin/ps --no-hea...pid,args -C perl': es2018.codfw.wmnet 100.0% (1/1) of nodes failed to execute command '/bin/ps --no-hea...pid,args -C perl': es2018.codfw.wmnet 0.0% (0/1) success ratio (< 100.0% threshold) for command: '/bin/ps --no-hea...pid,args -C perl'. Aborting. 0.0% (0/1) success ratio (< 100.0% threshold) for command: '/bin/ps --no-hea...pid,args -C perl'. Aborting. 0.0% (0/1) success ratio (< 100.0% threshold) of nodes successfully executed all commands. Aborting. 0.0% (0/1) success ratio (< 100.0% threshold) of nodes successfully executed all commands. Aborting. [WARNING]: Could not find a pt-heartbeat process to kill, using heartbeat table to determine the section Setting up original master as read-only Slave caught up to the master after waiting 0.298555850982666 seconds Servers sync at master: es2018-bin.002107:515019552 slave: es2017-bin.002103:916413841 Stopping original master->slave replication Setting up replica as read-write All commands where successful, current status: original master read_only: 1 / original slave read_only: 0 Trying to invert replication direction Starting heartbeat section b'es3' at es2017.codfw.wmnet ----- OUTPUT of '/usr/bin/nohup /...d &> /dev/null &' ----- ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 1.05hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/usr/bin/nohup /...d &> /dev/null &'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. ----- OUTPUT of '/bin/ps --no-hea...pid,args -C perl' ----- 4692 /usr/bin/perl /usr/local/bin/pt-heartbeat-wikimedia --defaults-file=/dev/null --user=root --host=localhost -D heartbeat --shard=es3 --datacenter=codfw --update --replace --interval=1 --set-vars=binlog_format=STATEMENT -S /run/mysqld/mysqld.sock --daemonize --pid /var/run/pt-heartbeat.pid 37349 /usr/bin/perl /usr/local/bin/pt-heartbeat-wikimedia --defaults-file=/root/.my.cnf -D heartbeat --shard=bes3 --datacenter=codfw --update --replace --interval=1 --set-vars=binlog_format=STATEMENT -S /run/mysqld/mysqld.sock --daemonize --pid /var/run/pt-heartbeat.pid ================ PASS: |████████████████████████████████████████████████████████████████████████████████| 100% (1/1) [00:00<00:00, 1.45hosts/s] FAIL: | | 0% (0/1) [00:00<?, ?hosts/s] 100.0% (1/1) success ratio (>= 100.0% threshold) for command: '/bin/ps --no-hea...pid,args -C perl'. 100.0% (1/1) success ratio (>= 100.0% threshold) of nodes successfully executed all commands. Detected heartbeat at es2017.codfw.wmnet running with PID 37349 Verifying everything went as expected... SUCCESS: Master switch completed successfully
Change 447648 had a related patch set uploaded (by Jcrespo; owner: Jcrespo):
[operations/software/wmfmariadbpy@master] mariadb: Fix heartbeat regex
Mentioned in SAL (#wikimedia-operations) [2018-07-24T16:58:00Z] <jynus> finishing test on es3 hosts T199224
Change 447648 merged by Jcrespo:
[operations/software/wmfmariadbpy@master] mariadb: Fix heartbeat regex
So the issue we had with pt-heartbeat was that the code ran by puppet wasn't actually running there- it doesn't get killed and restarted if it is running, with older parameters.
We need to kill heartbeat and restart it on all masters with the older paramaters. Alternatively, move heartbeat to a systemd-controled service.
Aside from this, I am happy with the testing done- I did a couple of extra runs to make sure everything was ok.
Thanks for the work you've put here. This is a nice work towards getting less downtime and less human errors while doing critical failovers.
Change 447659 had a related patch set uploaded (by Jcrespo; owner: Jcrespo):
[operations/software/wmfmariadbpy@master] switchover: Fix bug where shards are added with an extra 'b'
Change 447659 merged by Jcrespo:
[operations/software/wmfmariadbpy@master] switchover: Fix bug where shards are added with an extra 'b'