Alert re lkHWM near lkTableSize

Rob Fitzpatrick

Maybe this belongs in "Roadmap" rather than "Using", but anyway...

Looking at alert.cfg:

#
# Metric Type Compare Target Sensitivity Notify Message              Action
# ====== ==== ======= ====== =========== ====== ==================== =======================
# LogRd   num       > 100000         3:5 Always "Hit Ratio &1 &2 &3" alert-message,alert-log
#
# Metric        The ui-det name of the field being monitored.
#
# Type          Data type -- char or num (string or numeric).
#
# Compare       Operator  -- >, <, =, <>, <=, >=
#
# Target        The threshold value of the metric to be tested.

Since it isn't stated explicitly, does "Target" have to be a constant? Could it be another metric of the same data type as "Metric"? Or an expression on constants or metrics (e.g. lkHWM * 0.8)?

Use case: let's say I want to configure an alert to fire when the lock HWM is getting close to -L, and have it work regardless of the value of -L, which could possibly vary over time or from one DB to another. Is that possible? Or do I just need to decide which number (say, 7000) I care about and configure it to alert when lkHWM > 7000?

Food for thought: the alerting mechanism would become more powerful if the target values could be dynamic in some way. Maybe passed to the back end via the API.

tom

That is an interesting idea. A more complex rules engine for the alert processing is something that I have been thinking about for a while. On the other hand there is a lot to be said for keeping it simple.

Regarding the specific alert: there is a lkTblPct field which I use for a similar purpose. Rather than the HWM it reflects the locks in use at that moment. In some ways that is better -- the HWM sticks around until you restart and if something went bump in the night is going to keep alerting once you hit the threshold. Whereas lkTblPct will only alert if something is actively causing a problem.

Rob Fitzpatrick

@tom said in Alert re lkHWM near lkTableSize:

Regarding the specific alert: there is a lkTblPct field which I use for a similar purpose. Rather than the HWM it reflects the locks in use at that moment. In some ways that is better -- the HWM sticks around until you restart and if something went bump in the night is going to keep alerting once you hit the threshold.

Good to know, thanks.