I don't think so.
Afair you can cache the outputs of a whole task or the outputs of a whole artifact transform (of one artifact).
c
CristianGM
03/10/2022, 6:15 PM
With configuration-cache tasks are paralelizable, maybe it's worth using tasks instead of worker API
(Tricky, it only runs them in parallel if the configuration is cached)
v
Vampire
03/10/2022, 6:19 PM
If the tasks still use the worker api and don't explicitly wait for the result, they should also be run in parallel even without configuration cache enabled.
m
melix
03/10/2022, 7:12 PM
it isn't so obvious to run via tasks since new "work items" are discovered during execution. Using tasks I have to limit this to those I know upfront (which is already not trivial)
👍 1
l
lptr
03/11/2022, 3:19 PM
This is among the plans, but not supported right now.
lptr
03/11/2022, 3:20 PM
One reason for not doing this is the overhead each cached work item adds. If you have a task that runs 100 `WorkAction`s and each is cached separately, with the current cache protocol we’d be asking the cache server using 100 HTTP GET requests.
lptr
03/11/2022, 3:21 PM
But we want to implement worker action-level caching for native compilation, too.