[Swift-devel] Can we re-activate coaster worker timeout capability?

Michael Wilde wilde at mcs.anl.gov
Tue Nov 8 04:41:57 CST 2011


When using Swift with the OSG Glidein Workload Management System, which we need to do for the ExTENCI project, as well as with our own pilot job tools in bin/grid, we often will have more workers starting than we need.

It would be handy to re-instate some variation of the older worker timeout feature, under which when coasters is running in persistent passive mode, a worker option can specify a timeout period after which the worker will cleanly exit after some time period T of no work.

That way the pilot factory can aggressively launch workers, and if it overshoots, the excess workers will exit to avoid wasting site CPU resources.

Can that be done in a reasonable manner, avoiding the pitfalls that led to the removal of the timeout feature?

If so, I'll file an enhancement bug for this.

- Mike



More information about the Swift-devel mailing list