Re: Increasing delay of task-execution times


David Hudson
 

would declaring the tables help, or is using declares in the parent task just a way to speed  files opening in parent tasks so it doesnt have to open them over and over as the subtasks run.
I defer to the experts on that.

David Hudson
President GA Business Solutions
404-435-8630 

Meaningful data leads to better decisions.


On Mon, Oct 28, 2019 at 11:51 AM Avgerinos <mento@...> wrote:
Hi magicians

It's been 4 weeks that I 'm struggling with a weird behavior of XPA-runtime engine, noticed in all version versions from 3.3c and above, up 4.5a:

In a desktop app, there is a program executed by a timer event, once every second. This base-program is calling 20 resident sub-tasks (reading some data from database and making some calculations).
Some of the sub-tasks tasks read SQL2014 tables (on SSD drive), while the rest do only arithmetic calculations without accessing the database. 

During runtime, the whole process starts by taking around 350 milliseconds to execute, and this is quite acceptable for an event called every 1 second.
While executing the program keeps on slowing down: after 2 hours it will exceed the 1000 millisecond limit and will create an event-bottleneck.
Note that, this delay will not change if I stop and restart the event, or if I exit the program. The only way to reset back to normal execution times is to exit system and restart Magic.

While trying to locate the cause, I added time-logging code in all critical points. And this is where the strangest part comes:
  • Database accessing subtasks that were taking some time to execute (e.g. 15-ms), they still report the same execution time
  • Calculation subtasks with no access to database (exiting after a single record-suffix execution), report increasing execution times: they start from 0 ms and they gradually jump to 15-16 ms.
  • Note that these 15-16 ms seems to be the maximum accuracy I can get using the mTime() function for logging.
Some more remarks:
  • All subtasks are "non-abortable" (no events allowed)
  • I tried both options: resident or not resident tasks, and saw no difference
  • The behavior does not change whether executing from within the XPA-Studio (debug on or off), or directly with XPA-Runtime.
  • Tried many type of database locking schemes and saw no difference. Anyway the delay seems to happen more with non-database accessing subtasks
  • Tried replacing some of the SQL tables with memory tables, and saw no difference.
  • When tasks have already slowed down, the windows task-manager does not show any increase in CPU usage (always below 10%) or Memory usage (at least 40% of total memory free)

If anyone has similar experience or ideas, please let me know.

Thanks in advance

Avgerinos

Join {main@magicu-l.groups.io to automatically receive all group messages.