Debugging DRM

DRM Overview

 

•  Must READ:  DRM – Dynamic Resource management (Doc ID 390483.1)
• DRM cycle is very sensity – any problems like network problem, CPU starvation will lead to an instance/node reboot
• In 99% DRM is not the cause of the problem – it’s more a different problems occurs during DRM window
• LMON tracks the DRM cycle/steps
• Insert/Update/Delete operation will continue
• DRM cycles are often triggered by  hybrid workloads ( OLTP –> Batch -> OLTP … )

 Disable DRM and DRM config params

To disable DRM, set 
 _gc_affinity_time=0                           # Only if DB version is 10.1 or 10.2 
 _gc_undo_affinity=FALSE                       # Only if Db version is 10.2 
 _gc_policy_time=0                             # Only if DB version is 11.1 or higher
 _gc_affinity_time has been renamed to _gc_policy_time in 11g

DRM is driven by the following 
 1.) _gc_affinity_time = Time in minutes at which statistics will be evaluated (default = 10 mins) 
 2.) _gc_affinity_limit = # of times a node accesses a file/object (default = 50) 
 3.) _gc_affinity_minimum = minimum # of times per minute a file/object is accessed before affinity kicks in
 (default = 600 per minute per cpu )[This section is not visible to customers.]

Affinity Locks

•  Optimization introduced with 10.2
• Samller more efficient than Fusion Locks ( LE )
•  Affinity locks can be expanded to Fusion Locks and vice versa
• Affinity locks apply to data and undo segment blocks

  Affinity Locks sample

1. GCS lock ( LE ) mastered on instance 2
2. Instance 2 accesse buffer 50x more than instance 2
3. LEs dissolved and affinity lock created ( msstership stored in memory )
4. Instance 1 can now cheaply read/write to that buffer
5. Instance 2 access buffer again –> Affinity lock expanded to Fusion Lock ( LE )

Symptoms of a DRM problem

• DRM related wait event: gcs drm feeze in enter server mode
  •    –> Use Note: Script to Collect DRM Information (drmdiag.sql) (Doc ID 1492990.1)

• Wait Events with large buffer cache ( > 100 Gbyte )
  • remastering takes longer time more for large buffer caches
  • gcs remaster waits
  • DRM hang causes frequent RAC Instances Reconfiguration (Doc ID 1528362.1)

•   Database slowdown that correlate with DRMs
  •  –> Use Note: Script to Collect DRM Information (drmdiag.sql) (Doc ID 1492990.1)

•   DRM cycles must be finished in a certain time – if not instance crash with LMON timeout

Reference

• DRM – Dynamic Resource management (Doc ID 390483.1
• Script to Collect DRM Information (drmdiag.sql) (Doc ID 1492990.1)
• Bug 14641937 : HIGH WAIT TIME ON “GCS DRM FREEZE IN ENTER SERVER MODE”
• DRM hang causes frequent RAC Instances Reconfiguration (Doc ID 1528362.1)
• BUG:12879027 – LMON PROCESS CAN GET STUCK IN DRM QUIESCE STEP TRIGGERING PSEUDO RECONFIGURATION