Project

General

Profile

Actions

Bug #23192

open

Warn or error out if crunch-run cannot choose an interface address for local keepstore

Added by Tom Clegg 6 months ago. Updated 6 months ago.

Status:
New
Priority:
Normal
Assigned To:
-
Category:
Crunch
Target version:
-
Story points:
-
Release relationship:
Auto

Description

Currently, crunch-run tries to run local keepstore on a random port on a (non-loopback) local interface address, but if it can't figure out such an address for any reason, it runs keepstore on 0.0.0.0 and sets ARVADOS_KEEP_SERVICES=http://0.0.0.0:{port}/.

Although this works fine for arv-mount, saving logs, etc., this does not work if the container is running arvados-cwl-runner (or anything else that connects to Keep directly) -- the container process cannot connect to the host via 0.0.0.0:{port}.

(unless -container-network-mode=host, which it typically is not)

When this happens, and networking not disabled for the container, crunch-run should
  • Log an explanation that keepstore will be unreachable from the container because X (couldn't get list of network addresses / all configured addresses are loopback addresses)
  • Maybe error out -- or maybe not, since we can't be sure the container will actually depend on connecting to keepstore unless we see that the command is arvados-cwl-runner
Actions #2

Updated by Brett Smith 6 months ago

  • Release set to 82
Actions #3

Updated by Brett Smith 6 months ago

What if we just didn't set ARVADOS_KEEP_SERVICES for the container in this case? I think then the container would fall back to using the cluster's configured keepstores, which might be slower but seems most likely to keep everything working, right?

Actions

Also available in: Atom PDF