Closed andrewberls closed 9 years ago
:+1:
I think this probably justifies at least a minor version bump?
Theoretically speaking at least.
Yeah probably. I'm worried we'll reach 1.0 before it's actually a state I want to declare 1.0 though - given there have been 0 callers of fail
past or present I think this is ok as is
Haha I don't think it's a decimal system-- like after 0.9.0
is 0.10.0
, not 1.0.0
Not a blocker
No I mean I totally remember thinking the same thing.
Previously Overseer offered two flow-control constructs to users:
fail
andabort
. The idea was that aborts cause a job to halt immediately and cancel the execution of all its dependents, whereas a failure would not cancel dependent execution.In practice this does not make any sense - if a parent fails, a child will not be able to run, otherwise the dependency was incorrectly specified. Thus, this removes the failure construct in favor of supplanting the semantics of
abort
such thatabort
is now "loud abort", i.e. it has the same behavior but also logs errors and sends to Sentry, whereasabort-silent
also cancels dependents but does not send errors to Sentry.