dbendelman/dagster-testing

dagster-testing

To reproduce the issue, open 4 terminal windows on the project root and run each of these commands in a different terminal:

make run_good

make monitor_good

make run_bad

make monitor_bad

Then open 2 browser tabs, one for the working and one for the non-working instances.

Launch the bad_pipeline_parent pipeline in both the working and non working. Switch to the monitor_good and monitor_bad tabs and watch how in the good one the postgres connection count is normal, while in the bad one there is a connection leak. About 10 mins after launch, the bad instance's pipeline's bad_pipeline children will start erroring out with a Postgres "too many clients" error, while the good instance's continues to work fine indefinitely.

This issue is easy to spot in pgAdmin: X axis spans about 2 mins.

Binary commit search

Good 0794c34855b484d7d5ff4747c5b19764352ed040 (0.8.6) 72f2e129f6f3999d6930b86a148a9b2b6b70dfa6 jul 7 485365dba8393597e678815e6597b4031b3722da jul 8 27f04dc875572e8af809d24343ebc613914e4507 jul 9

90d404169d60fea714bdcd09b827c6de74c8a261 jul 9

Bad f711a5bce92f08c18f56fa20999366690c0b2788 jul 9 3f8e21850713062648e0d403b9a7fedb7aca2e93 jul 9 07f5cf80b224053afb8cbfbb6587da420dcc5dab (0.8.7)