Unreliable IE->SPARK round mapping

[bajtos]

The current implementation has several problems:

• We are running two instances of spark-api. They often get out of sync regarding the current SPARK round number. Depending on which instance a checker node reaches, it will receive tasks for the current or previous round.

• Recently, the Spark round 1909 lasted for less than a minute, and the Spark round 1914 lasted for almost an hour (two IE rounds).

Both problems can be seen on the following screenshot from Grafana. The first one happened around 4:00 am UTC.

[bajtos]

This is no longer an issue. spark-api reads the current round from the DB.