DeadlLocks may happen at LASTDATA POSTGRES SINK when upserting batches in HA scenarios

[mrutid]

Some deadlocks are arising at Postgres LastData Sink.

DEADLOCK
AGGREGATION-A UPSERT (-id:1,-id:...,-id:N)
AGGREGATION-B UPSERT (-id:N,-id:...,-id:1)

Sort aggregations by entity_id (at least) may help to avoid the problem.

NO DEADLOCK
AGGREGATION-A UPSERT (-id:1,-id:...,-id:N)
AGGREGATION-B UPSERT (-id:1,-id:...,-id:N)

Seems that each aggregation has the form:

{
   entity_id:  [id1,...idn]
   attr1:         [v1,..vn],
...
   attrM:       [v1,..vn]
}

There are multiple INSERT queries followed by a single Commit. We should order each query (by any common criteria).

https://github.com/telefonicaid/fiware-cygnus/blob/master/cygnus-common/src/main/java/com/telefonica/iot/cygnus/backends/sql/SQLBackendImpl.java#L610

This kind of snippet should make the work:

import java.util.ArrayList;
import java.util.Comparator;
upsertList.sort(Comparator.comparing(buff -> buff.toString()));

https://www.sololearn.com/compiler-playground/ceuF34hEdPlD

We should also change the attribute order within the query, so Time values, which is no key occurs later:

INSERT INTO watercycle_tank_lastdata (recvTime,fiwareServicePath,entityId,entityType,...)

to

INSERT INTO watercycle_tank_lastdata (entityId,entityType,fiwareServicePath recvTime,...)

fiware-cygnus/cygnus-ngsi/src/main/java/com/telefonica/iot/cygnus/aggregation/NGSIGenericColumnAggregator.java

Line 50 in aae3eb8

aggregation.put(NGSIConstants.RECV_TIME_TS+"C", new ArrayList<JsonElement>());

[fgalan]

PR #2199