Postgresql for mac buggy update#
Postgresql for mac buggy full#
Postgresql for mac buggy code#

This gist is a script that generates a minimal image which reproduces the bug.

The problematic image can be pulled from dockerhub: docker pull avivrosenberg/test-db-bug.

I donвЂ™t commit directly because then another layer is added and image size becomes a big issue. The reason IвЂ™m creating the second container and using docker cp (instead of just committing the first one) is that iвЂ™m actually creating multiple DB snapshots, each based on the previous one. The resulting image is the one that doesnвЂ™t run in Docker for mac.

docker commit the second container a new image.

Use docker cp to stream the PGDATA dir from the first container to the second, like so:Дocker cp $CONTAINER1: - | docker cp - $CONTAINER2:.

Create (but donвЂ™t start) a second postgres container, set custom PGDATA.

I say anecdotally because it just happened to us.

Stop the postgres process in the container so that everything is persisted to disk. Anecdotally, the Postgres replication process for keeping slaves up to date with the primary is a bit buggy.

This is optional, not necessary for reproducing the bug.

Write some data into the database using some program that connected to it.

Start a container based on postgres:9.5.3, with custom PGDATA.

So basically, the touch command I used failed for the same reason as the postgres process trying to create postmaster.pid in the PGDATA directory after changing permissions. Touch: cannot touch вЂ/var/lib/postgresql/test-data/foo.barвЂ™: Permission denied

Try to write to the PGDATA directory as the postgres user:.

ls -al $PGDATA, see that everything is now owned by postgres.

Change the permissions of PGDATA like they do in the official entrypoint script:Д¬hmod 700 '$PGDATA' & chown -R postgres '$PGDATA'.

ls -al $PGDATA, see that everything is owned by root.

Run a container based on the image with bash as entrypoint:Дocker run -rm -it -entrypoint=/bin/bash avivrosenberg/test-db-bug.

On Docker for mac, an error is displayed soon after starting the container:ДЇATAL: could not open lock file 'postmaster.pid': Permission denied Then the database loads without issue and is works fine (I can connect, read write data). Like I said, it runs fine on a linux version of docker.ДЄfter the entrypoint script is executed, the permissions of PGDATA are correctly set to the postgres user. Mkdir -p '$PGDATA' chmod 700 '$PGDATA' chown -R postgres '$PGDATA' Expected behavior This is not an issue normally, because the postgres official image comes with an entrypoint script that starts with the following: The owner of the custom PGDATA directory is root (because it was copied in).

Postgresql for mac buggy full#

See вЂњSteps to reproduceвЂќ for a full description of how the image was created. The purpose of said image is to serve as a pre-seeded DB for tests running in a CI server.Дuring tests, I want to start a DB container that contains quite a lot of data, allow the test to read and write to it, and then after the test I want all the modifications discarded so they donвЂ™t affect the next test that uses this image.Д«asically the image uses a non-volume directory as PGDATA, containing data that was copied into the container with docker cp. This feels like a serious bug since the premise of docker is that an image, once created, should run exactly the same everywhereвЂ¦ Information This image runs fine on docker in linux (tested with ver 1.11.2), but fails to run on Docker for Mac beta (1.12.0-rc4-beta20, diagnostic id = 100BE54F-2248-4431-9382-6BA42F9FA120). Note that HLL is optimised for pre-aggregated counts, it isn't terribly fast at doing adhoc count estimates over millions of rows (although still faster than a raw COUNT DISTINCT).I have created an Image based on the official postgres:9.5.3 image. If it's not faster than HLL then I'll just be using HLL. But I can use extensions that are available in AWS RDS postgres.

Postgresql for mac buggy code#

failing that, are there any binary data structures that allow mutable/in-place set_bit edits?ДЄ constraint is that I can't push C code or extensions to implement this.

Postgresql for mac buggy update#

is there a way to inplace / mutably update a BIT or bytea?.However, it is very slow because I am using set_bit on a BIT and that has copy-on-write semantics. ) over a UUID column, a kind of poor man's HyperLogLog (and having different perf characteristics). I have written an AGGREGATE function that approximates a SELECT COUNT(DISTINCT.