Evol Monkey

At 9.5 pg_rewind was introduced, with this feature it was possible to  make a server that is no longer master to follow a promoted standby that has a new timeline.  There are at least a few use cases that could benefit from this feature, to name a few : Testing Failovers : Its pretty common for one of customers to ask for a failover simulation just to make sure that everything works, by using pg_rewind this is much easier and much faster because we don't have to re-sync the whole database that sometimes is multi-TB sized.  Minor version upgrades : during an upgrade you have to put the new binaries in place and restart the database. By using pg_rewind if the application is flexible enough , you could upgrade the slave , switchover , upgrade the master and pg_rewind the old master to follow the new slave, possibly minimising even more the downtime. Requirements. pg_rewind has a couple of requirements : 1. data page checksums enabled ( initdb -k ) or...

I recently came across a discussion in #postgresql channel about how to add a new column that will represent record insert timestamps to a table with minimal locking. Normally you would do something like this : monkey=# alter table alter_test add column insert_timestamp timestamp without time zone default now(); ALTER TABLE Time: 13643.983 ms monkey=# create index concurrently insert_timestamp_idx on alter_test (insert_timestamp) where insert_timestamp is not null; CREATE INDEX Time: 1616.108 ms This though , would require an AccessExclusiveLock on the table and the application would have to wait for the duration of alter. What you could do instead is : monkey=# alter table alter_test add column insert_timestamp timestamp without time zone; ALTER TABLE Time: 1.135 ms monkey=# alter table alter_test alter column insert_timestamp set default now(); ALTER TABLE Time: 1.238 ms monkey=# create index concurrently insert_timestamp_idx on alter_test (insert_timestamp)...

Lately, i wanted to make a new test / lab environment that would be dispensable, fast to deploy and easily customisable. VM's are all nice for lab use and they served me well over the years, but they tend to be slow(er) to deploy, heavy on resources and lets face it, So i've decided to play around with Docker. I've played with Docker in the past but i haven't done anything too serious, i just took the official postgres Dockerfile and run it to see how it works. This is how i started my whole Docker exploration, i took the official postgres docker file that can be found here  and made an image. Immediately i saw that it wasn't exactly what i wanted, i wanted something less generic. The things that i would like are : latest postgres version, compiled from source easily customisable pgdata and prefix custom postgresql.conf settings (mostly having it replication ready) custom shell environment some extensions data page checksums easy replica deployment...