DKB@ATLAS instance
Current status: OK.
ToDo:
- Check/update AMI client version (viktor)
- Check master vs. data4es-prod (mgolosova)
- (If item 2 is OK) Reset data4es-prod on master and apply the new version in production (viktor)
Integration w/ BigPanDA Mon
Study report on the "time series issue":
- Task metadata from DEFT give us not pure "time series", but "cumulative sums" series -- with gaps.
- Cumulative sums with gaps cannot be aggregated over multiple series as-is.
- ES derivative aggregation cannot correctly reconstruct time series from cumsums (gaps are treated incorrectly).
- InfluxDB correctly works with gaps, but the whole query that will generate the goal series (for whole campaign) is not constructed yet.
- Theoretical study of InfluxDB technical characteristics say that it can work with "ephemeral" time series (multiple short series instead of few indefinitely continuous series).
ToDo:
- This week:
- full InfluxDB query for campaign step statistics;
- storage scheme;
- data load module (stage);
- prepare the real-life data sample;
- load sample to InfluxDB@bamboo instance;
- Next steps:
- InfluxDB instance installation/configuration via Puppet;
- pre-load T-stage (after 040?);
- adjustment of the data4es process;
- run new version in production;
- REST API server method.
Paperwork
Suggestion: publish ex-ICCS paper and DKB new-instance-guide as preprints.
ToDo:
- Check if KI preprint is an option.
- ...