Closed mmesim closed 2 years ago
Thanks for the report.
I saw this error many times
13:58:47 [error] [detector-02::8b18b121-6d0a-4a80-86f9-f8f8e84491aa::aa538897-997e-4219-971c-eb3a9e89a256] [8D.RAW2..HHN] Failed to feed data (tw.start=2019-11-05T01:14:32.455Z, tw.end=2019-11-05T01:14:34.355Z). Reason: status=101, status_value=0.000000 (failed to feed data)
What's the configured waveformBufferSize
? Is there data for the required waveform stream identifier (i.e. 8D.RAW2..HHN
) and time window available?
Any clue? Is the VM or the code?
There is a memory issue during the application clean-up. Could you provide data including the configuration in order to reproduce the issue? Thanks.
What's the configured
waveformBufferSize
? Is there data for the required waveform stream identifier (i.e.8D.RAW2..HHN
) and time window available?
I didn't set any attribute waveformBufferSize
, so I guess is the default (?).
I just found it in scconfig
:open_mouth:
It was 300 seconds. I think a value of 60 seconds would be ok, right?
It was 300 seconds. I think a value of 60 seconds would be ok, right?
scautopick
's default is 300 s (see https://docs.gempa.de/seiscomp/current/apps/scautopick.html#confval-ringBufferSize). For the config you run this shouldn't be an issue (though, less is better in order to reduce scdetect-cc
's memory footprint). Note that the required data for amplitude calculation still must be buffered. Else amplitude calculation (and therefore magnitude calculation, too) fails.
@mmesim, I had a closer look into the issue.
First of all, the memory issue should be fixed with #100.
Besides, I had a look into the issue causing the error log entry
13:58:47 [error] [detector-02::8b18b121-6d0a-4a80-86f9-f8f8e84491aa::aa538897-997e-4219-971c-eb3a9e89a256] [8D.RAW2..HHN] Failed to feed data (tw.start=2019-11-05T01:14:32.455Z, tw.end=2019-11-05T01:14:34.355Z). Reason: status=101, status_value=0.000000 (failed to feed data)
In order to understand what's going on behind the scenes it is important to remind that data required for amplitude calculation must be still buffered. Though, if trigger facilities are enabled the current detection is emitted (and therefore processed) only once the next detection is available. However, if the time passed by (i.e. the time between the current detection and the next one) was greater than the time specified by processing.waveformBufferSize
module configuration parameter no buffered data is available anymore and thus amplitude calculation fails. As a result you see an error log entry as e.g. above. With #100 I handled this case more gracefully. Though, the fact that there is no amplitude calculated cannot be avoided. AFAIU, the issue can be partially avoided only if fetching historical data is implemented (e.g. similar to the approach by scamp
).
EDIT: Although, it's just a workaround, increasing the waveformBufferSize
should help. Again, note that this will increase scdetect-cc
's memory footprint.
Well, I was trying to run
scdetect-cc
as :When running the same dataset without requiring magnitude calculation, I was able to get results, i.e 3330 origins -> 1109 unique events.
With the magnitude on
I saw this error many times
At some point it crushed
Any clue? Is the VM or the code?