drmemtrace scheduler turns all-blocked runqueue into fatal STATUS_IDLE error #7318
Assignees
Labels
Comments
derekbruening
added a commit
that referenced
this issue
Mar 4, 2025
Rebalancing was propagating an idle status from a runqueue to the caller, where it was treated as an error when it is in fact innocuous. Adds a unit test that reproduces the error without the fix and passes with the fix. Fixes #7318
This was referenced Mar 4, 2025
derekbruening
added a commit
that referenced
this issue
Mar 4, 2025
Rebalancing was propagating an idle status from a runqueue to the caller, where it was treated as an error when it is in fact innocuous. Adds a unit test that reproduces the error without the fix and passes with the fix. Fixes #7318
derekbruening
added a commit
that referenced
this issue
Mar 12, 2025
If an input is bound to every output, we now ignore those bindings, which eliminates complexities in initial output allocation where rebalancing is required to even things out (and that gets complicated by initially-blocked inputs) as well as removes overhead in the runqueue code during dynamic scheduling. Binding to every output is actually not uncommon as users have such bindings as a default value. As part of detecting bind-to-all, invalid bindings are now detected. Adds unit tests for the invalid bindings. Updates the unbalanced rebalance test to bind to all but one as binding to all is no longer unbalanced. Issue: #7318
derekbruening
added a commit
that referenced
this issue
Mar 13, 2025
If an input is bound to every output, we now ignore those bindings, which eliminates complexities in initial output allocation where rebalancing is required to even things out (and that gets complicated by initially-blocked inputs) as well as removes overhead in the runqueue code during dynamic scheduling. Binding to every output is actually not uncommon as users have such bindings as a default value. As part of detecting bind-to-all, invalid bindings are now detected. Adds unit tests for the invalid bindings. Updates the unbalanced rebalance test to bind to all but one as binding to all is no longer unbalanced. Issue: #7318
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
If during a rebalance an output's runqueue has only blocked inputs, it will return STATUS_IDLE, which the scheduler propagates as a fatal error even though it should be innocuous. It shows up as this message:
The text was updated successfully, but these errors were encountered: