[HDDS-9879] Memory leak handling block/container token in OM/SCM

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.4.0
• Labels: pull-request-available
• Assignee: Sumit Agrawal
• Reporter: George Huang
• Created: 08/Dec/23 05:23
• Updated: 12/Dec/23 17:56
• Resolved: 10/Dec/23 08:09

Description

OOM is observed for FIXED_BYTESTRING_CACHE consuming approx 22GB of memory. 

 
This is called for token converting to proto for: Kind (fixed) and service (its either container id and blockId), so service part calling below just adding to map consuming memory.
 
setKindBytes(getFixedByteString(tok.getKind())).
setServiceBytes(getFixedByteString(tok.getService()));
 
org.apache.hadoop.ozone.protocolPB.OzonePBHelper#getFixedByteString
public static ByteString getFixedByteString(Text key) {
ByteString value = FIXED_BYTESTRING_CACHE.get(key);
if (value == null)

{ value = ByteString.copyFromUtf8(key.toString()); ByteString oldValue = FIXED_BYTESTRING_CACHE.putIfAbsent(key, value); return oldValue != null ? oldValue : value; }

return value;
}
 
 Token service is either containerId or blockId, so no need maintain cache for this as most of time, this will be unique.

Issue Links

• relates to: HDDS-9882 Share ByteString-related code between OzonePBHelper classes
• links to: GitHub Pull Request #5748