You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
An obscure bug where old tasks aren't being removed from the scheduler data store after drifting out of the n=1 window, reproduced by Dave.
There is a reproducible example for this, but it's rather odd and very sensitive to change.
This workflow will restart itself in cycle 2, then pause itself in the last cycle. The task 1/c will remain in the data store incorrectly.
You can observe this in cylc tui (which uses a global update, no deltas) or cylc dump.
[scheduling]
cycling mode = integer
initial cycle point = 1
final cycle point = 6
[[graph]]
P1 = """ b[-P1] => a => b => c => d => e e[-P1] => e"""[runtime]
[[a, b, c, e]]
script = sleep 5
[[b]]
script = """ if [[ $CYLC_WORKFLOW_FINAL_CYCLE_POINT == $CYLC_TASK_CYCLE_POINT ]]; then cylc pause $CYLC_SUITE_NAME elif [[ $CYLC_TASK_CYCLE_POINT == "2" ]]; then sleep 15 cylc stop --now $CYLC_SUITE_NAME sleep 15 cylc play --host=localhost $CYLC_SUITE_NAME else sleep 30 fi"""
[[d]]
script=sleep 20
The good news is that I have not been able to reproduce this on #5660 which indicates that the issue is within the housekeeping of the increment_graph_window algorithm which is good news as a fix is on the horizon.
The text was updated successfully, but these errors were encountered:
An obscure bug where old tasks aren't being removed from the scheduler data store after drifting out of the n=1 window, reproduced by Dave.
There is a reproducible example for this, but it's rather odd and very sensitive to change.
This workflow will restart itself in cycle 2, then pause itself in the last cycle. The task
1/c
will remain in the data store incorrectly.You can observe this in
cylc tui
(which uses a global update, no deltas) orcylc dump
.The good news is that I have not been able to reproduce this on #5660 which indicates that the issue is within the housekeeping of the
increment_graph_window
algorithm which is good news as a fix is on the horizon.The text was updated successfully, but these errors were encountered: