Occasional corruption of polled_cmd property
Occasionally, usually after we perform an upgrade/redeploy, we see some devices with a mangled polled_cmd property.
These are SKA devices where the only polled command is ExecutePendingOperations
. I understand that these devices set the period to 5ms on startup, and then after init to 100ms.
Last week when I reported this on Mattermost, it was a device server hosting four devices, where the symptom was a segfault on startup:
- device 1 was correct
- device 2 had the value
["ExecutePendingOperations"]
- device 3 had the value
["100", "100"]
- device 4 had the value
["100"]
This has just re-occurred with a different device server hosting a single device, whose polled_cmd
property looked like ["ExecutePendingOperations", "5", "100"]
. This device started apparently normally, but then displayed odd behaviour - not sure how much is relevant to Tango Controls code as opposed to SKA however.
I am not able to reliably reproduce this problem (but manually manging the property value reliably produces the bad behaviour).
As well as a fix for whatever bug is causing the corruption, I suppose a check that these special attributes are in the right format on device startup would make this problem easier to detect.