Answer Diff [0]
<<toc>>
¶
== General Principles ¶
¶
Recovery is tested via the RQG, which provides a random workload against the ¶
server, and then uses kill -9 to kill the process. After that, recovery is ¶
attempted both by using ##maria_read_log## and by restarting the ##mysqld## ¶
process. Once the server has started up, the tables are verified in various ¶
ways, including ##ALTER|OPTIMIZE|ANALYZE|REPAIR TABLE## as well ##SELECT## ¶
queries that read the table back and forth using various access methods. ¶
¶
A combinations ##.CC## file named ¶
##lp:randgen/conf/engines/maria/maria_recovery.cc## is used to define various ¶
##mysql## options and RQG parameters that are relavant to recovery. Then, RQG's ¶
##combinations.pl## script is used to run hundreds of individual test runs. ¶
Each run uses a random permutation from the settings in the ##.CC## file in ¶
order to generate a unique workload that is then validated via the ##Recovery## ¶
RQG Reporter. ¶
¶
== Individual tests ¶
¶
The following are the individual tests or test runs that must be completed or ¶
created in order to ensure that Aria recovery is solid. ¶
¶
=== Standard kill -9 testing ¶
¶
**Done 2011-02-28** The standard ##conf/engines/maria/maria_recovery.cc## ¶
passes with no failures when run with hundreds of trials. ¶
¶
=== Testing with small block sizes ¶
¶
On hold pending 2 bug fixes related to --maria-block-size=1K and ¶
--maria-block-size=2K ¶
¶
=== Testing with small page cache size ¶
¶
**Done 2011-03-04** Completed 400 rounds with ¶
¶
<<code>> ¶
'--mysqld=--maria-block-size=4K --mysqld=--maria-pagecache-buffer-size=128K', ¶
'--mysqld=--maria-block-size=16K --mysqld=--maria-pagecache-buffer-size=256K', ¶
'--mysqld=--maria-block-size=32K --mysqld=--maria-pagecache-buffer-size=512K' ¶
<</code>> ¶
¶
two pre-recovery crashes were filed, no recovery issues. ¶
¶
=== Killing and restarting the recovery process itself ¶
¶
**In Progress** ¶
¶
The AriaDoubleRecovery reporter currently attempts doble recovery via ¶
maria_read_log. The first invocation of maria_read_log is killed halfway ¶
through the process and the second invocation is left to complete the recovery. ¶
¶
Future testing will involve doing the same with the mysqld server in place of ¶
maria_read_log. ¶
¶
=== Another realistic workload ¶
¶
The usefullness of the SMF workload, derived from the SimpleMachines forum ¶
application means that another such workload is required in order to make sure ¶
no residual recovery bugs remain. Hopefully something can be cooked up using ¶
Wikipedia so that longer fields and blobs are exercised. ¶
¶
=== Transactional consistency ¶
¶
A transactional grammar that simulates transactions using LOCK TABLEs is ¶
required. The RecoveryConsistency Reporter can then be used to validate that no ¶
partial transactions appear in the database after recovery. ¶
¶
== General Principles ¶
¶
Recovery is tested via the RQG, which provides a random workload against the ¶
server, and then uses kill -9 to kill the process. After that, recovery is ¶
attempted both by using ##maria_read_log## and by restarting the ##mysqld## ¶
process. Once the server has started up, the tables are verified in various ¶
ways, including ##ALTER|OPTIMIZE|ANALYZE|REPAIR TABLE## as well ##SELECT## ¶
queries that read the table back and forth using various access methods. ¶
¶
A combinations ##.CC## file named ¶
##lp:randgen/conf/engines/maria/maria_recovery.cc## is used to define various ¶
##mysql## options and RQG parameters that are relavant to recovery. Then, RQG's ¶
##combinations.pl## script is used to run hundreds of individual test runs. ¶
Each run uses a random permutation from the settings in the ##.CC## file in ¶
order to generate a unique workload that is then validated via the ##Recovery## ¶
RQG Reporter. ¶
¶
== Individual tests ¶
¶
The following are the individual tests or test runs that must be completed or ¶
created in order to ensure that Aria recovery is solid. ¶
¶
=== Standard kill -9 testing ¶
¶
**Done 2011-02-28** The standard ##conf/engines/maria/maria_recovery.cc## ¶
passes with no failures when run with hundreds of trials. ¶
¶
=== Testing with small block sizes ¶
¶
On hold pending 2 bug fixes related to --maria-block-size=1K and ¶
--maria-block-size=2K ¶
¶
=== Testing with small page cache size ¶
¶
**Done 2011-03-04** Completed 400 rounds with ¶
¶
<<code>> ¶
'--mysqld=--maria-block-size=4K --mysqld=--maria-pagecache-buffer-size=128K', ¶
'--mysqld=--maria-block-size=16K --mysqld=--maria-pagecache-buffer-size=256K', ¶
'--mysqld=--maria-block-size=32K --mysqld=--maria-pagecache-buffer-size=512K' ¶
<</code>> ¶
¶
two pre-recovery crashes were filed, no recovery issues. ¶
¶
=== Killing and restarting the recovery process itself ¶
¶
**In Progress** ¶
¶
The AriaDoubleRecovery reporter currently attempts doble recovery via ¶
maria_read_log. The first invocation of maria_read_log is killed halfway ¶
through the process and the second invocation is left to complete the recovery. ¶
¶
Future testing will involve doing the same with the mysqld server in place of ¶
maria_read_log. ¶
¶
=== Another realistic workload ¶
¶
The usefullness of the SMF workload, derived from the SimpleMachines forum ¶
application means that another such workload is required in order to make sure ¶
no residual recovery bugs remain. Hopefully something can be cooked up using ¶
Wikipedia so that longer fields and blobs are exercised. ¶
¶
=== Transactional consistency ¶
¶
A transactional grammar that simulates transactions using LOCK TABLEs is ¶
required. The RecoveryConsistency Reporter can then be used to validate that no ¶
partial transactions appear in the database after recovery. ¶
¶
