So I had this situation when cluster started with 8 nodes and 2 went down for some reason.
As a user, my only way to detect this situation:
- query failed because something started to execute on a node and failed because it went down (and for that I have to comb through the logs to find a warning)
- my queries are extremely slow, because my queries started to execute after node went down and got deregistered from zookeeper.
- somebody just stopped drillbit on a particular node
Since there is no central place (apart from zookeeper) where information on participating nodes is kept, when I queried sys.drillbits, I got 6 nodes, as if 2 others never existed ...There is beauty in flexibilty, but in real life situation when you have more than 20 nodes, things can get out control quickly.
Since zookeeper has this information in the first place, can we enhance sys.drillbits table to have drillbit status as zookeeper sees it ?
This can also help with testing and automating test cases that test for failure conditions like that.