nomad-dispatch-repro

README

nomad dispatch job tester

Following dispatch jobs is harder than it ought to be. This is a test repro based on real code to demonstrate how difficult it is to get right.

This code will register a parameterized job and then dispatch many instances of it: attempting to follow each dispatch to its completion.

prerequisites

1. A running Nomad cluster (can be nomad agent -dev)
2. NOMAD_ADDR, NOMAD_TOKEN, and any other env vars necessary to register jobs set.
3. Go

running

# If you have the source checked out:
$ go run .
2024/11/14 15:56:06 [ 0:0   ] ok
2024/11/14 15:56:06 [ 0:0   ] ok
2024/11/14 15:56:06 [ 0:0   ] ok
2024/11/14 15:56:06 [ 1:0   ] ok
2024/11/14 15:56:06 [ 1:0   ] expected "sleeper" task but none is found for dispatch ID sleeper/dispatch-1731628560-bbbe6f03
... snip ...
2024/11/14 15:57:12 [ 2:9   ] ok
2024/11/14 15:57:12 [ 2:9   ] ok
2024/11/14 15:57:12 500 done after 1m12.399s with 53 errors

For certain errors the allocation(s) in question will be written to the current directory as json such as: feb895ce-2aaa-8460-c754-cf05a8127370.alloc.json

Use -help to see and adjust test parameters.

The test does not cleanup after itself, but only creates 1 job (and a number of dead allocs).

example output

With no errors as of 215630b2a785114d77bb648f24a17ee5bd0b1a64:

2024/11/14 16:09:00 500 done after 1m9.374s with 0 errors