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
I was on Mastodon trumpeting the fact that Humility mostly works with my keyboard when I noticed...a party pooper.
It's subtle, but: see how scanner has its timer set for T+10?
scanner always sets its timer for T+1. It sleeps for one millisecond at a time.
So why would Humility show it with a timer set for T+10? Because Humility captures the system time early in the process (see where it's printed up top) and uses that value to compare to task timers. The T+10 means that it's resumed the target for 9 ms during the task snapshot process, generating an inconsistent snapshot.
I can reproduce inconsistent snapshots here pretty reliably -- for instance, getting output where no tasks are running despite being runnable (the idle task, at least, should show as running!).
This suggests to me that the issue where we're not reliably halting during snapshots has regressed. 😢
The text was updated successfully, but these errors were encountered:
This is causing me significant pain trying to debug the sidecar startup VPD read issue right now. It would be nice to see which tasks are blocking on which during the 11.6 second delay, but the dumps show random things blocked in impossible positions because they're not all captured at the same time.
I was on Mastodon trumpeting the fact that Humility mostly works with my keyboard when I noticed...a party pooper.
It's subtle, but: see how
scanner
has its timer set forT+10
?scanner
always sets its timer forT+1
. It sleeps for one millisecond at a time.So why would Humility show it with a timer set for T+10? Because Humility captures the system time early in the process (see where it's printed up top) and uses that value to compare to task timers. The T+10 means that it's resumed the target for 9 ms during the task snapshot process, generating an inconsistent snapshot.
I can reproduce inconsistent snapshots here pretty reliably -- for instance, getting output where no tasks are running despite being runnable (the idle task, at least, should show as running!).
This suggests to me that the issue where we're not reliably halting during snapshots has regressed. 😢
The text was updated successfully, but these errors were encountered: