[HBASE-27219] Change JONI encoding in RegexStringComparator

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 2.5.0, 3.0.0-alpha-4, 2.4.14
• Component/s: Filters
• Hadoop Flags: Reviewed

Release Note

In RegexStringComparator an infinite loop can occur if an invalid UTF8 is encountered. We now use joni's NonStrictUTF8Encoding instead of UTF8Encoding to avoid the issue.

• Assignee: Minwoo Kang
• Reporter: Minwoo Kang
• Created: 18/Jul/22 07:07
• Updated: 21/Jul/22 02:13
• Resolved: 19/Jul/22 16:23

Description

I change the engine of RegexStringComparator to JONI.
After that I sent a regex filter request, the RegionServer's heap memory usage spiked and the RegionServer did not work due to GC.
 

(RegionServer Heap Memory Usage)
 

INFO  [JvmPauseMonitor] util.JvmPauseMonitor: Detected pause in JVM or host machine (eg GC): pause of approximately 1435ms
GC pool 'ParNew' had collection(s): count=1 time=1550ms
INFO  [JvmPauseMonitor] util.JvmPauseMonitor: Detected pause in JVM or host machine (eg GC): pause of approximately 1073ms
GC pool 'ParNew' had collection(s): count=1 time=1534ms
INFO  [JvmPauseMonitor] util.JvmPauseMonitor: Detected pause in JVM or host machine (eg GC): pause of approximately 1456ms
GC pool 'ParNew' had collection(s): count=1 time=1574ms
INFO  [JvmPauseMonitor] util.JvmPauseMonitor: Detected pause in JVM or host machine (eg GC): pause of approximately 1297ms
GC pool 'ParNew' had collection(s): count=1 time=1415ms 

(RegionServer Log)
 
I checked the reason, it is said that when using UTF8Encoding, an infinite loop can occur if an invalid UTF8 is entered.
For trino, using NonStrictUTF8Encoding instead of UTF8Encoding.
(https://github.com/trinodb/trino/commit/ea66e8cb27b098a5cea184106fe245064351b567)

After changing the encoding of JoniRegexEngine to NonStrictUTF8Encoding in RegexStringComparator, it was confirmed that the heap memory usage spike was gone.

Issue Links

• links to: GitHub Pull Request #4632