Sequence – SV contention
Table of Contents
Waits for cached sequence when sequence is used from multiple instance
10046 trace: WAIT #139899759861648: nam='enq: SV - contention' ela= 4276 type|mode=1398145029 id1=88851 id2=0 obj#=-1 tim=1385221394056801 SQL> select owner, object_name, object_type from dba_objects where object_id=88851; OWNER OBJECT_NAME OBJECT_TYPE ----------------------- ---------------------- ------------------ SCOTT T1_SEQ SEQUENCE
Waits for UNCACHED sequences when used from multiple instances
WAIT #140557198270352: nam='row cache lock' ela= 1227 cache id=13 mode=0 request=5 obj#=-1 tim=1384249119100670
SQL> select type, parameter from v$rowcache where cache#=13;
TYPE PARAMETER
----------- --------------------------------
PARENT dc_sequences
..
WAIT #140557198260504: nam='gc cr block 2-way' ela= 529 p1=1 p2=667 p3=1 obj#=74 tim=1384249119101498
WAIT #140557198253360: nam='gc current block 2-way' ela= 687 p1=1 p2=667 p3=33554433 obj#=74 tim=1384249119102947
SQL> SELECT segment_name, segment_type, block_id, blocks
FROM dba_extents
WHERE file_id = &file_no AND ( &block_value BETWEEN block_id
AND ( block_id + blocks -1 ) );
Enter value for file_no: 1
Enter value for block_value: 667
SEGMENT_NAME SEGMENT_TYPE BLOCK_ID BLOCKS
------------- -------------------- ---------- ----------
SEQ$ TABLE 664 8a
- For each seq.nextval we are waiting 1.2 ms for a row cache lock and 0.6 ms for a CR request on SEQ$ and again 0.6 ms for getting a the most CURRENT block from SEQ$.
- The complete wait time using a UNCACHED sequence is about 2.4 ms for each seq.nextval operation
- UNCACHED sequences triggers a library cache lock where CACHED sequences triggers a SV lock
Performance test results
Test Env : 11.2.0.4 3-node RAC 3-node Rac cluster / 10 Threads each inserting 1000 rows - 10.000 rows in summary Test Command Runtime TPS Using sequence with ORDER NOCACHE $ java UCPDemo grac4 10 1000 -seq 46s 212 Using sequence with ORDER CACHE 20 $ java UCPDemo grac4 10 1000 -seq 33s 297 Using sequence with ORDER CACHE 1000 $ java UCPDemo grac4 10 1000 -seq 32s 312 Using programming-methods to create $ java UCPDemo grac4 10 1000 -noseq 11s 875
- Always try to use application based sequences if possible ( about 4x faster than ORDER NOCACHE )
- Use CACHE attribute if possible and avoid ORDER NOCACHE combination
- For heavily used sequences use 1000 for its cache size – for infrequently use sequences use 20 for cache size parameter
AWR report for NOCACHE ORDER sequence sample
--------------------------------------------------- ---------------------- ------------------------------ -------------------------- I# Class Event Waits %Timeouts Total(s) Avg(ms) %DB time Avg Min Max Std Dev Cnt ---- ---------- ------------------------- ------------ --------- ------------- ------- -------- -------- -------- -------- -------- ---- * Concurrenc row cache lock 9,572 0.0 368.76 38.5 82.70 38.26 35.73 42.92 4.04 3 DB CPU N/A N/A 25.39 N/A 5.69 3 Commit log file sync 9,968 0.0 24.44 2.5 5.48 2.42 2.07 2.69 0.32 3 Other row cache process 73,768 0.0 5.78 0.1 1.30 0.08 0.03 0.17 0.08 3 Cluster gc cr block busy 2,896 0.0 4.37 1.5 0.98 1.66 1.28 2.37 0.61 3 Cluster gc current block 2-way 6,603 0.0 4.08 0.6 0.91 0.62 0.60 0.64 0.02 3 Cluster gc current block 3-way 3,223 0.0 3.44 1.1 0.77 1.08 0.94 1.32 0.20 3 Cluster gc cr block 2-way 4,047 0.0 3.18 0.8 0.71 0.83 0.66 0.99 0.17 3 Cluster gc cr block 3-way 2,431 0.0 2.87 1.2 0.64 1.43 1.12 1.59 0.27 3 Other gcs log flush sync 3,310 0.0 1.87 0.6 0.42 0.53 0.49 0.59 0.05 3
AWR report for CACHE 1000 ORDER sequence sample
---------------------------------------- ---------------------- ------------------------------ ----------------------------------- I# Class Event Waits %Timeouts Total(s) Avg(ms) %DB time Avg Min Max Std Dev Cnt ---- ---------- ----------------------------- ------------ --------- ------------- ------- -------- -------- -------- -------- -------- * Other enq: SV - contention 9,953 0.0 197.99 19.9 80.68 21.22 7.87 28.11 11.56 3 Commit log file sync 10,005 0.0 24.83 2.5 10.12 2.41 2.04 3.09 0.59 3 DB CPU N/A N/A 12.25 N/A 4.99 3 Configurat enq: HW - contention 73 0.0 3.12 42.7 1.27 43.72 34.21 59.78 13.99 3 Other reliable message 5 0.0 1.79 357.3 0.73 594.44 1.67 1779.91 1026.65 3 Cluster gc current block 3-way 250 0.0 1.58 6.3 0.64 5.18 1.91 11.56 5.52 3 Concurrenc library cache lock 30 10.0 0.96 32.1 0.39 34.92 25.53 52.26 15.04 3 Cluster gc buffer busy acquire 118 0.0 0.94 8.0 0.38 8.64 3.69 13.70 5.01 3 Cluster gc current block 2-way 612 0.0 0.86 1.4 0.35 1.41 0.78 2.54 0.98 3 Cluster gc current block busy 161 0.0 0.81 5.0 0.33 4.96 3.92 5.66 0.92 3
- Each update on a sequence requires a row cache lock ( Caching reduces this updates – but can lead to gaps if instance crashes )
- if using ORDER and NOCACHE attribute every seq.nextval access triggers a row cache lock
- if using NOCACHE sequence row cache lock event could be a major wait event
- If using CACHE enq: SV – contention becomes a major wait event