Have MemoryTracker report VmHWM
I finally asked the right question and receive the answer that it is the Resident Set Size (
RSS) memory that the grid monitors to chastise too demanding jobs.
In old code,
VmPEAK was reported because I thought that was the relevant figure. It turns out, the relevant figure is instead
VmHWM (High Water Mark).
I would like
MemoryTracker to report that value (printing it after each module is bonus points).
There is a
getrusage() call (
sys/resource.h) that is supposed to give that value and to work in both OSX and Linux. I could not obtain meaningful results from it (or I am not sure they were meaningful), but there it is.
/proc/PID/status reports that value as
/proc/PID/stat does not.
#2 Updated by Kyle Knoepfel over 3 years ago
- Status changed from Accepted to Resolved
- Assignee set to Kyle Knoepfel
- Target version set to 2.01.00
- % Done changed from 0 to 100
- SSI Package art added
- SSI Package deleted (
The RSS high-water mark for the entire process is now reported in the
MemoryTracker general summary (e.g.):
MemoryTracker General SUMMARY (all numbers in units of Mbytes) Peak virtual memory usage (VmPeak) : 503.145 Mbytes Peak resident set size usage (VmHWM): 147.055 Mbytes
While we are able to implement more fine-grained HWM reporting--e.g. after each event/module--doing so has larger consequences, perhaps involving an adjustment to the database schema. If desired, we request that you open a separate feature request for it.
General-summary feature implemented with art:ae6441a0.