Markus, are you involved with / on the pgsql-hackers-cluster mailing list?
On Wed, 7 Apr 2010, Markus Schmidleitner wrote:
hi list, hi mitani! @all here's another patch based on the patches i sent in before. the current patch is against postgresql 8.3.10. i hope anyone will give it a try. comments/suggestions are highly appreciated! @mitani maybe you can do some code review/testing on it? would be nice ;-) i also attached a small script and config files which i use for regression testing. you may need to change some paths etc. how to build? 1. download postgresql-8.3.10.tar.gz 2. extract postgresql-8.3.10.tar.gz 3. mv postgresql-8.3.10 to pgcluster 4. cd pgcluster 5. patch -p1 < (path_to_patch_file)/postgresql-8.3.10-to-pgcluster-1.9.14.patch 7. mkdir obj 8. cd obj 9. ../configure && make maintainer-clean 10. ../configure && make ... what's new? - reworked replication (now the work is done locally first and then replicated - so queries which will fail anyway don't get replicated any more) - reworked hot-recovery - reworked PREPAREd statement handling - include SAVEPOINT handling - reject queries on cluster if it is used for recovery - fixed accept bug in pgreplicate - fixed bugs in pgc_compare - handle errors from remote host's so that query will fail on source too - treat processes lifecheck, recovery provider and recovery as auxilary processes (autorestart, shutdown, ...) - start lifecheck, recovery provider ... only if not in standalone mode - implemented cluster connect/shutdown message - some other improvements/fixes/changes changes from earlier patches: - made regression tests working again (in replication mode) - default regression tests (make test) now run in standalone-mode omitting all replication code - implemented cluster-compare utility (pgc_compare) - reworked replication of copy commands - fixed bug in scan_terminate - fixed some other bugs/issues that prevented regression tests to run sucessfully - usage of transactions in conjunction with AutoLockTable now works - some refactorings and code cleanups - initdb now works as expected (failed in previous patches) - added postgresql startup parameter '-a' to set the cluster in full standalone mode, bypassing all replication operations (mainly for regression testing and an operational mode which we need here) - added a few configuration parameters (see current cluster.conf and pgreplicate.conf) - now you can specify a comma separated list of hostnames or addresses in Host_Name-Parameter (currently this is limited to a set of 2 - see cluster.conf and pgreplicate.conf) - pgreplicate and postgresql recovery processes now listen only on those interfaces that are specified - set default-value of auto_lock_table to false - reworked locking code and deadlock notification - some refinements on networking-code - use modified elog instead of own logging-code (mainly for logging in common code used by pgreplicate and postgresql - currently logging from pgreplicate to a logfile is not implemented) - some more general reworks, dead code eliminations ... - replication server will be noticed to all clusters - cluster will be noticed to all replication servers - standby replication server send cascade info to cluster * cluster will automatically switch to primary replication server - primary replication server requests replication servers from cluster * replication server will automatically setup cascade * clusters automatically switch back to primary replication server - added threading-support in pgreplicate (mainly backported from cybercluster) - code refactoring * implemented functions for non-blocking i/o which are used by entire pgcluster * all net-i/o operations are non-blocking * earse dead code * remove needless memcpy * other minor reworks - implemented a cluster-notice message when cascading is used * it informs the upper and lower cascade that cluster finished recovery and can be used now best regards, Markus Schmidleitner
---- Marc G. Fournier Hub.Org Hosting Solutions S.A. [email protected] http://www.hub.org Yahoo:yscrappy Skype: hub.org ICQ:7615664 MSN:[email protected] _______________________________________________ Pgcluster-general mailing list [email protected] http://pgfoundry.org/mailman/listinfo/pgcluster-general
