[HBASE-10424] HMaster could capture stacks of RSes it deems unresponsive during assignments

• Type: Wish
• Status: Closed
• Priority: Trivial
• Resolution: Won't Fix
• Affects Version/s: 0.96.0
• Component/s: Region Assignment
• Assignee: Unassigned
• Reporter: Harsh J
• Created: 27/Jan/14 10:46
• Updated: 17/Jun/22 05:03
• Resolved: 08/Jun/14 21:24

Description

Often there are cases of a region not getting assigned due to timeouts (while others do go through). In this case, the Master does appear to enter a never-ending retry operation where it retries each chosen server several times before moving to another.

For debugging in such a scenario, where the master is best aware of the situation, it could use that to its advantage and help capture issues better if it probably setup an N retry threshold (for # of servers tried) and run a HTTP GET on the current timing out RS's info port, to capture its /stacks end point and dump the output in its logs for investigation later.