Our FR10 trend display began stalling after I enabled detailed logging, especially when the folder is busy. The cell controls are separate. I suspect synchronous writes, but I want a cheap timing check before changing storage or blaming communications.
Measure elapsed time around acquisition, parsing, log writing and display delivery separately, using a monotonic clock. Keep that timing evidence away from the same synchronous log path under investigation, or the measurement itself may queue behind the delay you need to see.
The existing timestamps are written only after the whole update returns. They show a gap, not which stage consumed it. I can ask our developer to add those separate measurements.
Sofia, good point. If one slow write blocks the next poll from starting, the communication call can be fast and the displayed sample still be old. I'd want sample age visible as well as the duration of the latest call.
Yes, distinguish scheduling delay from acquisition duration. For a local comparison, replay the same captured input with normal and deliberately slow test storage. Keep the acquisition test separate so that replay does not get presented as a finding about the network.
Developer added the measurements to the replay harness. With the slow test sink, time accumulates in the synchronous writer and subsequent updates start late; parsing stays short. This reproduces the display stall without a robot connection.
That's a useful reproduction, Fiona. Has the proposed fix got a bounded backlog and an explicit response when storage stays slow? Moving the write elsewhere can make the screen look healthy while the logger accumulates an expensive secret.
We have a candidate worker with a bounded queue, plus an overflow warning and dropped-diagnostic count. In the slow-sink replay the display keeps updating and the warning appears; we have not yet compared it with the workshop acquisition path.
Keep the dropped count separate from missing acquisition samples. A logging overflow should not be reported as a communication loss unless you actually observed one.
Agreed. The candidate names it dropped diagnostic records, and shows sample age separately. Replay fix looks useful; communication timing in the workshop remains an open check, not a conclusion from the local test.