Perilous New Undocumented Tracing Feature in Oracle 21c
oradebug request completion
revealed DBWR wait event "oradebug request completion" and made an in-depth investigation.
This Blog will demonstrate the reproducing of this event and explore further internals.
Note: Tested on Oracle 19.27 with:
db_writer_processes = 3
disk_asynch_io = true
aio-max-nr = 1048576 (for 1048576/4096 = 256 Oracle processes in Linux)
1. Test Setup
1.1 Tablespace, table, index
We create a wide table of 100 rows on a slow file system with each data block containing one single row.
-- Create a file system of 10MB (20480 512-bytes sectors) with 5000 ms read/write delay
-- [root@test_db3] > echo "0 20480 delay /dev/loop3 0 5000" | dmsetup create dm-slow3
-- [root@test_db3] > mount /dev/mapper/dm-slow3 /home/oracle/suk/slow_dir3
drop tablespace test_tbs_slow3 including contents and datafiles cascade constraints;
create SMALLFILE tablespace test_tbs_slow3 datafile
'/home/oracle/suk/slow_dir3/test_tbs_slow3.dbf' size 5M autoextend on next 1M maxsize 9M
logging online extent management local autoallocate blocksize 8k
segment space management auto flashback on;
-- Elapsed: 00:02:28.63
drop table test_tab_slow3 cascade constraints;
create table test_tab_slow3(id number, text1 varchar2(3510), text2 varchar2(3510))
tablespace test_tbs_slow3 storage (initial 16k next 128k);
create index test_tab_slow3#id on test_tab_slow3(id) tablespace test_tbs_slow3;
insert into test_tab_slow3 select level id, rpad('ABC', 3000, 'X') text1, rpad('ABC', 3000, 'X') text2 from dual connect by level <= 100;
commit;
1.2 Procedures
Then create job and "flush buffer_cache" procedures:
create or replace procedure update_row3(p_job_id number, p_loop_count number) as
begin
for i in 1..p_loop_count loop
update test_tab_slow3 set text1 = rpad('ABC', 3500, 'X')||1, text2 = rpad('ABC', 3500, 'X')||i where id between (p_job_id-1)*10 and p_job_id*10-1;
--dbms_session.sleep(0.1);
commit;
--execute immediate 'alter system checkpoint';
end loop;
end;
/
-- exec update_row3(1, 1);
create or replace procedure update_row_jobs3(p_job_count number, p_loop_count number) as
begin
for i in 1..p_job_count loop
dbms_scheduler.create_job (
job_name => 'TEST_JOB3_'||i,
job_type => 'PLSQL_BLOCK',
job_action => 'begin update_row3('||i||', '||p_loop_count||'); end;',
start_date => systimestamp,
--repeat_interval => 'systimestamp',
auto_drop => true,
enabled => true);
end loop;
end;
/
--exec update_row_jobs3(10, 10000);
--clearup_test
create or replace procedure flush_bc (p_cnt number) as
begin
for i in 1..p_cnt loop
execute immediate 'alter system flush buffer_cache';
end loop;
end;
/
-- exec flush_bc(1000);
1.3 Slow I/O Test
Run following code with Sql Trace:
alter system flush buffer_cache;
alter session set tracefile_identifier = 'slow_trc_1';
alter session set events '10046 trace name context forever, level 12';
select * from test_tab_slow3 where id = 3;
alter session set events '10046 trace name context off';
*** 2025-08-28T10:47:16.022188+02:00
WAIT #140226820024472: nam='db file sequential read' ela= 5176667 file#=2607 block#=139 blocks=1 obj#=5938335 tim=8556907488219
*** 2025-08-28T10:47:21.141086+02:00
WAIT #140226820024472: nam='db file sequential read' ela= 5118663 file#=2607 block#=133 blocks=1 obj#=5938334 tim=8556912607124
SQL ID: c6b234hzgd4u9 Plan Hash: 745759405
select * from test_tab_slow3 where id = 3
call count cpu elapsed disk query current rows
------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 1 0.00 0.00 0 0 0 0
Fetch 2 0.00 10.29 2 3 0 1
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 4 0.00 10.29 2 3 0 1
Rows (1st) Row Source Operation
---------- ---------------------------------------------------
1 TABLE ACCESS BY INDEX ROWID BATCHED TEST_TAB_SLOW3 (cr=3 pr=2 pw=0 time=10295880 us starts=1 cost=1 size=337428 card=84)
1 INDEX RANGE SCAN TEST_TAB_SLOW3#ID (cr=2 pr=1 pw=0 time=5177141 us starts=1 cost=1 size=0 card=34)(object id 5938335)
Elapsed times include waiting on following events:
Event waited on Times Max. Wait Total Waited
---------------------------------------- Waited ---------- ------------
SQL*Net message to client 2 0.00 0.00
Disk file operations I/O 1 0.00 0.00
db file sequential read 2 5.17 10.29
SQL*Net message from client 2 0.07 0.07
********************************************************************************
2. Test Run
2.1 Test Start
We launch 10 table update JOBs and then "flush buffer_cache" to force DB Writers to actively write data to disk.
declare
l_job_cnt number := 10;
l_running_job_cnt number := 0;
begin
update_row_jobs3(l_job_cnt, 100000);
while l_running_job_cnt < l_job_cnt loop
select count(*) into l_running_job_cnt from dba_scheduler_jobs where job_name like '%TEST_JOB%';
dbms_session.sleep(1);
end loop;
dbms_session.sleep(4);
flush_bc(1000);
end;
/
2.2 DBW0 SIGUSR2 signal Catching
We compose one GDB script and use it to catch SIGUSR2 on DBW0:
------------- gdb_dbw0_script_1.txt -------------
set pagination off
set logging file gdb_dbw0_script_1.log
set logging overwrite on
set logging on
set debug timestamp
set $sig_cnt = 0
# we can also use: catch signal SIGUSR2
break sigprocmask
commands
shell date
printf "===== sigprocmask (%i)=====\n", ++$sig_cnt
bt 8
shell sleep 30
cont
end
cont
$ > gdb -x gdb_dbw0_script_1.txt -p 604093
Breakpoint 1 at 0x7fdfa1303630
Program received signal SIGUSR2, User defined signal 2.
0x0000000001f72ce9 in kcbblwr ()
(gdb) c
Continuing.
Breakpoint 1, 0x00007fdfa1303630 in sigprocmask () from /lib64/libc.so.6
Tue Aug 26 20:52:15 CEST 2025
===== sigprocmask (1)=====
#0 0x00007fdfa1303630 in sigprocmask () from /lib64/libc.so.6
#1 0x00000000035e2900 in sspuser ()
#2
#3 0x0000000001f72ce9 in kcbblwr ()
#4 0x0000000001f5e5bd in kcbbdrv ()
#5 0x00000000137cdf55 in ksb_act_run_int ()
#6 0x00000000137cce04 in ksb_act_run ()
#7 0x000000000407b4be in ksbabs ()
Breakpoint 1, 0x00007fdfa1303630 in sigprocmask () from /lib64/libc.so.6
Tue Aug 26 20:52:45 CEST 2025
===== sigprocmask (2)=====
#0 0x00007fdfa1303630 in sigprocmask () from /lib64/libc.so.6
#1 0x00000000035e296d in sspuser ()
#2
#3 0x0000000001f72ce9 in kcbblwr ()
#4 0x0000000001f5e5bd in kcbbdrv ()
#5 0x00000000137cdf55 in ksb_act_run_int ()
#6 0x00000000137cce04 in ksb_act_run ()
#7 0x000000000407b4be in ksbabs ()
2.3 DB alert.log Watching
When DBW0 (session ID 810 serial# 24481 OSPID 604093) receives SIGUSR2, DB hangs and becomes unresponsive with DBW0 as Final blocker.
Other DB writers (DBW1 and DBW2) shows "executing after 5 secs due to non-DBWR task".
$ > tail -f alert_test_db3.log
2025-08-26T20:38:35.835071+02:00
DIA0 Critical Database Process As Root: Hang ID 9 blocks 10 sessions
Final blocker is session ID 810 serial# 24481 OSPID 604093 on Instance 1
If resolvable, instance eviction will be attempted by Hang Manager
2025-08-26T20:38:40.848025+02:00
System State dumped to trace file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_dia0_604091_base_1.trc
2025-08-26T20:39:23.354993+02:00
System State dumped to trace file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_dia0_604091_lws_2.trc
2025-08-26T20:39:45.434724+02:00
System State dumped to trace file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_dia0_604091_lws_2.trc
2025-08-26T20:40:24.557175+02:00
System state dump requested by (instance=1, osid=604097 (DBW1)), summary=[system state dump request (ksudss)].
System State dumped to trace file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_diag_604074_20250826204024.trc
2025-08-26T20:40:24.561550+02:00
DBW 1 executing after 5 secs due to non-DBWR task at ts=1756233624
2025-08-26T20:40:33.925273+02:00
DBW 2 executing after 5 secs due to non-DBWR task at ts=1756233633
2025-08-26T20:46:13.941192+02:00
2.4 bpftrace tracing tgkill (sender, receiver and signal)
We trace tgkill (sender, receiver, signal) with following script.
A comment like "# DIA0 -> DBW0" is added in the output to denote "DIA0 sends SIGUSR2 (sig: 12) to DBW0".
(In the output, we only show one full callstack of first event).
"# DIA0 -> DBW0" is triggered by "oradebug short_stack" from DIA0 on DBW0.
If DBW0 is slow or there are a huge write, DIA0 session shows Wait Event: "oradebug request completion".
If it takes more than 5 seconds, v$system_event.total_timeouts increased 1.
We can see different pairs of (sender, receiver) like "DBW1 -> CKPT", "DBW0 -> DBW2".
It looks like some automatic diagnostic activities among Oracle processes.
$> bpftrace -e 'BEGIN {@seq = 0}
uprobe:/orabin/app/oracle/product/19.27.0.0.250415-212/bin/oracle:sskgp_tgkill+2
{
time("%H:%M:%S --- ");
printf("%d --- comm=%s pid=%d rsi(receiver_pid)=%d rdx(sig)=%d \n", @seq++, comm, pid, reg("si"), reg("dx"));
printf("%s\n", ustack())
}'
Attaching 2 probes...
20:37:22 --- 65 --- comm=ora_dia0_test_d pid=604091 rsi(receiver_pid)=604093 rdx(sig)=12 # DIA0 -> DBW0
sskgp_tgkill+2
ksdx_cmdreq_issue+644
ksdx_cmdreq_reissue_attempt+433
ksdx_cmdreq_exec_track_debugee+517
ksdx_cmdreq_exec+173
ksdxdocmdmultex+1768
ksdxdocmdmult+55
ksudmp_proc_short_stack+373
kjznlwshngsew+2725
kjznlwstrigdmps+268
kjznlwschksess+1303
kjznrmnscanrmnsess+11216
kjznrmnhmdetect+283
kjznrmnhm+27
kjznhm+1892
kjzdregf+906
kjzdm+691
ksbrdp+1167
opirip+541
opidrv+581
sou2o+165
opimai_real+173
ssthrdmain+417
main+256
__libc_start_main+243
0x5541f689495641d7
20:37:37 --- 66 --- comm=ora_dia0_test_d pid=604091 rsi(receiver_pid)=620881 rdx(sig)=12 # DIA0 -> J008
20:40:19 --- 74 --- comm=ora_dbw1_test_d pid=604097 rsi(receiver_pid)=604109 rdx(sig)=12 # DBW1 -> CKPT
20:40:55 --- 94 --- comm=ora_diag_test_d pid=604074 rsi(receiver_pid)=604097 rdx(sig)=12 # DIAG -> DBW1
...
20:51:53 --- 167 --- comm=ora_mmon_test_d pid=604139 rsi(receiver_pid)=604043 rdx(sig)=12 # MMON -> PMON
...
20:53:09 --- 184 --- comm=ora_dia0_test_d pid=604091 rsi(receiver_pid)=604139 rdx(sig)=12 # DIA0 -> MMON
20:53:29 --- 186 --- comm=ora_dbw0_test_d pid=604093 rsi(receiver_pid)=604097 rdx(sig)=12 # DBW0 -> DBW1
20:53:30 --- 187 --- comm=ora_dbw0_test_d pid=604093 rsi(receiver_pid)=604101 rdx(sig)=12 # DBW0 -> DBW2
20:53:31 --- 202 --- comm=ora_diag_test_d pid=604074 rsi(receiver_pid)=604091 rdx(sig)=12 # DIAG -> DIA0
2.5 bpftrace on DB Writers
We use following script to trace DBWR signal catching (only full callstack are showed for first two events).
It shows DBWR receive signals.
[root@test_db3] >bpftrace -e 'BEGIN {@seq = 0}
tracepoint:syscalls:sys_enter_rt_sigprocmask /strncmp("ora_dbw", comm, 7) == 0/
{
time("%H:%M:%S --- ");
printf("%d --- comm=%s pid=%d \n", @seq++, comm, pid);
printf("%s\n", ustack())
}'
Attaching 2 probes...
19:53:53 --- 0 --- comm=ora_dbw1_test_d pid=604097
__GI___sigprocmask+17
0x7f80476f1c20
skgpwwait+187
ksliwat+2293
kslwaitctx+200
ksarcv+376
ksbabs+670
ksbrdp+1167
opirip+541
opidrv+581
sou2o+165
opimai_real+173
ssthrdmain+417
main+256
__libc_start_main+243
0x5541f689495641d7
19:53:53 --- 1 --- comm=ora_dbw1_test_d pid=604097
pthread_sigmask+74
ksdxcb+499
sspuser+223
0x7f80476f1c20
skgpwwait+187
ksliwat+2293
kslwaitctx+200
ksarcv+376
ksbabs+670
ksbrdp+1167
opirip+541
opidrv+581
sou2o+165
opimai_real+173
ssthrdmain+417
main+256
__libc_start_main+243
0x5541f689495641d7
20:40:24 --- 3 --- comm=ora_dbw2_test_d pid=604101
__GI___sigprocmask+17
20:40:30 --- 4 --- comm=ora_dbw2_test_d pid=604101
pthread_sigmask+74
20:52:45 --- 15 --- comm=ora_dbw0_test_d pid=604093
__GI___sigprocmask+17
20:52:50 --- 16 --- comm=ora_dbw0_test_d pid=604093
pthread_sigmask+74
2.6 wait_event "oradebug request completion" systemstate dump
During test, we also set event to generate a system-state trace when a wait of bigger than 10ms on 'oradebug request completion' is detected:
ALTER SYSTEM SET events = 'wait_event["oradebug request completion"]{wait: minwait=10000} systemstate(266)';
--ALTER SYSTEM SET events = 'wait_event["oradebug request completion"]{wait: minwait=10000} off';
SQL> oradebug setmypid
Statement processed.
SQL> oradebug eventdump system
wait_event ["oradebug request completion"] {wait: minwait=10000} systemstate(266)
System State dumped to trace file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_diag_604074_20250826204024.trc
--------------- test_db3_diag_604074_20250826204024.trc ---------------
Unix process pid: 604074, image: oracle@test_db3 (DIAG)
*** 2025-08-26T20:40:24.557417+02:00
System state caller stack:
Short stack dump:
ksedsts()+426<-ksdxfstk()+63<-ksdxdocmdmultex()+5627<-ksdxdocmdmult()+55<-ksudmp_proc_short_stack()+373<-ksudss_opt()+307<-kjzddss()+545<-kjzdmdrq()+6930<-kjzdregf()+4183<-kjzdm()+
691<-ksbrdp()+1167<-opirip()+541<-opidrv()+581<-sou2o()+165<-opimai_real()+173<-ssthrdmain()+417<-main()+256<-__libc_start_main()+243<-_start()+46
===================================================
SYSTEM STATE (level=10, with short stacks)
------------
Dumping the Process Summary
8: USER ospid 619494 sid 26 ser 53200, waiting for 'DLM cross inst call completion'
Cmd: PL/SQL EXECUTE
Blocked by inst: 1, sid: 811, ser: 4896
Final Blocker inst: 1, sid: 811, ser: 4896
11: DIAG ospid 604074 sid 808 ser 30814,
18: DIA0 ospid 604091 sid 541 ser 16758, waiting for 'DIAG idle wait'
19: DBW0 ospid 604093 sid 810 ser 24481,
20: DBW1 ospid 604097 sid 5 ser 1428, waiting for 'oradebug request completion'
21: DBW2 ospid 604101 sid 273 ser 22218, waiting for 'rdbms ipc message'
22: LGWR ospid 604105 sid 542 ser 2191, waiting for 'rdbms ipc message'
23: CKPT ospid 604109 sid 811 ser 4896, waiting for 'rdbms ipc message'
24: LG00 ospid 604113 sid 6 ser 7534, waiting for 'LGWR worker group idle'
25: SMON ospid 604115 sid 274 ser 52613, waiting for 'enq: CR - block range reuse ckpt'
Blocked by inst: 1, sid: 811, ser: 4896
Final Blocker inst: 1, sid: 811, ser: 4896
26: LG01 ospid 604119 sid 543 ser 61931, waiting for 'LGWR worker group idle'
56: J000 ospid 620865 sid 21 ser 42342, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPDATE
Blocked by inst: 1, sid: 810, ser: 24481
Final Blocker inst: 1, sid: 810, ser: 24481
57: J001 ospid 620867 sid 292 ser 4060, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPDATE
Blocked by inst: 1, sid: 810, ser: 24481
Final Blocker inst: 1, sid: 810, ser: 24481
58: J002 ospid 620869 sid 560 ser 27312, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPDATE
Blocked by inst: 1, sid: 810, ser: 24481
Final Blocker inst: 1, sid: 810, ser: 24481
66: J00A ospid 620994 sid 561 ser 35509, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPDATE
Blocked by inst: 1, sid: 810, ser: 24481
Final Blocker inst: 1, sid: 810, ser: 24481
68: M002 ospid 618975 sid 22 ser 30742, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPDATE
Blocked by inst: 1, sid: 810, ser: 24481
Final Blocker inst: 1, sid: 810, ser: 24481
69: MZ00 ospid 620998 sid 295 ser 41723, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPSERT
Blocked by inst: 1, sid: 810, ser: 24481
Final Blocker inst: 1, sid: 810, ser: 24481
70: M004 ospid 618953 sid 557 ser 23163, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPSERT
Blocked by inst: 1, sid: 810, ser: 24481
Final Blocker inst: 1, sid: 810, ser: 24481
2025-08-25T10:50:32.827370+02:00
DIA0 Critical Database Process As Root: Hang ID 1 blocks 11 sessions
Final blocker is session ID 810 serial# 31793 OSPID 601523 on Instance 1
If resolvable, instance eviction will be attempted by Hang Manager
2025-08-25T10:50:37.839966+02:00
System State dumped to trace file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_dia0_601521_base_1.trc
2025-08-25T10:50:39.601999+02:00
Errors in file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_dia0_601521_base_1.trc:
ORA-32515: cannot issue ORADEBUG command 'SHORT_STACK' to process 'Unix process pid: 601523, image: oracle@test_db3 (DBW0)'; prior command execution time exceeds 5000 ms
2025-08-25T10:50:44.283104+02:00
DIA0 Critical Database Process As Root: Hang ID 2 blocks 3 sessions
Final blocker is session ID 811 serial# 20147 OSPID 601539 on Instance 1
If resolvable, instance eviction will be attempted by Hang Manager
2025-08-25T10:52:10.789009+02:00
System state dump requested by (instance=1, osid=601531 (DBW2)), summary=[system state dump request (ksudss)].
2025-08-25T10:52:10.795339+02:00
Errors in file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_dbw2_601531.trc:
ORA-32515: cannot issue ORADEBUG command 'SHORT_STACK' to process 'Unix process pid: 601523, image: oracle@test_db3 (DBW0)'; prior command execution time exceeds 5000 ms
2025-08-25T10:52:10.805672+02:00
Errors in file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_dbw1_601527.trc:
ORA-32515: cannot issue ORADEBUG command 'SHORT_STACK' to process 'Unix process pid: 601523, image: oracle@test_db3 (DBW0)'; prior command execution time exceeds 5000 ms
System State dumped to trace file /orabin/app/oracle/admin/test_db3/diag/rdbms/test_db3/test_db3/trace/test_db3_diag_601504_20250825105210.trc
2025-08-25T10:52:11.546767+02:00
DBW 1 executing after 6 secs due to non-DBWR task at ts=1756111931
2025-08-25T10:52:11.646360+02:00
DBW 2 executing after 6 secs due to non-DBWR task at ts=1756111931
ORA-32515: cannot issue ORADEBUG command 'SHORT_STACK' to process 'Unix process pid: 601523, image: oracle@test_db3 (DBW0)'; prior command execution time exceeds 5000 ms
0x0000000003357d4b <+2715>: mov $0x1388,%edx
0x0000000003357d50 <+2720>: callq 0x13570f0
----------- test_db3_diag_601504_20250825105210.trc -----------
19: DBW0 ospid 601523 sid 810 ser 31793, waiting for 'db file parallel write'
20: DBW1 ospid 601527 sid 5 ser 51041,
21: DBW2 ospid 601531 sid 273 ser 53811,
22: LGWR ospid 601535 sid 542 ser 33370, waiting for 'rdbms ipc message'
23: CKPT ospid 601539 sid 811 ser 20147, waiting for 'rdbms ipc message'
34: MMON ospid 601569 sid 545 ser 42846, waiting for 'log file switch (checkpoint incomplete)'
Blocked by inst: 1, sid: 810, ser: 31793
Final Blocker inst: 1, sid: 810, ser: 31793
38: USER ospid 601876 sid 554 ser 30768, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPDATE
Blocked by inst: 1, sid: 810, ser: 31793
Final Blocker inst: 1, sid: 810, ser: 31793
47: W003 ospid 601811 sid 816 ser 37737, waiting for 'log file switch (checkpoint incomplete)'
Blocked by inst: 1, sid: 810, ser: 31793
Final Blocker inst: 1, sid: 810, ser: 31793
57: USER ospid 601960 sid 290 ser 217, waiting for 'DLM cross inst call completion'
Cmd: PL/SQL EXECUTE
Blocked by inst: 1, sid: 811, ser: 20147
Final Blocker inst: 1, sid: 811, ser: 20147
58: M002 ospid 601840 sid 555 ser 44546, waiting for 'class slave wait'
59: J002 ospid 601942 sid 824 ser 55525, waiting for 'log file switch (checkpoint incomplete)'
Cmd: UPDATE
Blocked by inst: 1, sid: 810, ser: 31793
Final Blocker inst: 1, sid: 810, ser: 31793
60: J000 ospid 601938 sid 20 ser 28041, waiting for 'enq: US - contention'
Cmd: UPDATE
Blocked by inst: 1, sid: 826, ser: 1291
Final Blocker inst: 1, sid: 810, ser: 31793
2.7 Session 'oradebug request completion' Stats
select localtimestamp, event, total_waits, total_timeouts, round(time_waited_micro/total_waits) ms_per_wait, round(time_waited_micro/1e6) wait_seconds --, v.*
from v$system_event v where lower(event) = 'oradebug request completion';
LOCALTIMESTAMP EVENT TOTAL_WAITS TOTAL_TIMEOUTS MS_PER_WAIT WAIT_SECONDS
---------------------- ---------------------------- ----------- -------------- ----------- ------------
27-AUG-2025 12:14:45 oradebug request completion 2507 67 1006642 2524
select sample_time, program, session_id, session_serial#, event
from v$active_session_history t
where sample_time between timestamp'2025-8-26 20:38:40' and timestamp'2025-8-26 20:40:35'
and event in ('oradebug request completion')
order by t.sample_time;
SAMPLE_TIME PROGRAM SESSION_ID SESSION_SERIAL# EVENT
----------------------- ------------------------ ---------- --------------- ---------------------------
26-AUG-2025 20:38:40 oracle@test_db3 (DIA0) 541 16758 oradebug request completion
26-AUG-2025 20:40:19 oracle@test_db3 (DBW1) 5 1428 oradebug request completion
26-AUG-2025 20:40:20 oracle@test_db3 (DBW1) 5 1428 oradebug request completion
26-AUG-2025 20:40:21 oracle@test_db3 (DBW1) 5 1428 oradebug request completion
26-AUG-2025 20:40:22 oracle@test_db3 (DBW1) 5 1428 oradebug request completion
26-AUG-2025 20:40:23 oracle@test_db3 (DBW1) 5 1428 oradebug request completion
26-AUG-2025 20:40:24 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:25 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:26 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:27 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:28 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:29 oracle@test_db3 (DBW2) 273 22218 oradebug request completion
26-AUG-2025 20:40:29 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:30 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:30 oracle@test_db3 (DBW2) 273 22218 oradebug request completion
26-AUG-2025 20:40:31 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:31 oracle@test_db3 (DBW2) 273 22218 oradebug request completion
26-AUG-2025 20:40:32 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:32 oracle@test_db3 (DBW2) 273 22218 oradebug request completion
26-AUG-2025 20:40:33 oracle@test_db3 (DBW2) 273 22218 oradebug request completion
26-AUG-2025 20:40:33 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
26-AUG-2025 20:40:34 oracle@test_db3 (DIAG) 808 30814 oradebug request completion
2.8 Session Process Info
Here the session info ( SID, SERIAL#, PID, SPID) in our test:
select s.program, event, s.sid, s.serial#, pid, spid
from v$session s, v$process p
where s.paddr=p.addr and (lower(s.program) like '%(dia%' or lower(s.program) like '%(dbw%'
or lower(s.program) like '%(ckpt%' or lower(s.program) like '%(lg%' or lower(s.program) like '%(j0%');
PROGRAM EVENT SID SERIAL# PID SPID
------------------------- -------------------------- ----- -------- ----- ------
oracle@test_db3 (DIAG) DIAG idle wait 808 30814 11 604074
oracle@test_db3 (DIA0) DIAG idle wait 541 16758 18 604091
oracle@test_db3 (DBW0) db file parallel write 810 24481 19 604093
oracle@test_db3 (DBW1) db file parallel write 5 1428 20 604097
oracle@test_db3 (DBW2) db file parallel write 273 22218 21 604101
oracle@test_db3 (LGWR) rdbms ipc message 542 2191 22 604105
oracle@test_db3 (CKPT) rdbms ipc message 811 4896 23 604109
oracle@test_db3 (LG00) LGWR worker group idle 6 7534 24 604113
oracle@test_db3 (LG01) LGWR worker group idle 543 61931 26 604119
oracle@test_db3 (J000) db file sequential read 21 42342 56 620865
oracle@test_db3 (J001) free buffer waits 292 4060 57 620867
oracle@test_db3 (J002) db file sequential read 560 27312 58 620869
oracle@test_db3 (J003) free buffer waits 829 13992 59 620871
oracle@test_db3 (J004) db file sequential read 20 48258 60 620873
oracle@test_db3 (J005) db file sequential read 289 7069 61 620875
oracle@test_db3 (J006) db file sequential read 562 16826 62 620877
oracle@test_db3 (J007) db file sequential read 830 19629 63 620879
oracle@test_db3 (J008) db file sequential read 2 61229 64 620881
oracle@test_db3 (J009) db file sequential read 291 8814 65 620883
oracle@test_db3 (J00A) jobq slave wait 561 24036 66 620885
3. Observing "oradebug request completion"
With following small test, we can watch "oradebug request completion". Open two Sqlplus sessions (PID 36 and 59):
select s.last_call_et, s.program, event, s.sid, s.serial#, pid, spid
from v$session s, v$process p
where s.paddr=p.addr and lower(s.program) like '%sqlplus%' order by 1;
LAST_CALL_ET PROGRAM EVENT SID SERIAL# PID SPID
------------ ------------------------------ ---------------------------- ---- -------- ---- ------
55 sqlplus.exe SQL*Net message from client 19 50486 36 613512
639 sqlplus@test_db3 (TNS V1-V3) SQL*Net message from client 829 46358 59 623458
$> gdb -p 613512
(gdb) catch signal SIGUSR2
Catchpoint 1 (signal SIGUSR2)
(gdb) handle SIGUSR2
Signal Stop Print Pass to program Description
SIGUSR2 Yes Yes Yes User defined signal 2
(gdb) c
SQL> oradebug setorapid 36
Oracle pid: 36, Unix process pid: 613512, image: oracle@test_db3
SQL> oradebug short_stack
(gdb) catch signal SIGUSR2
Catchpoint 1 (signal SIGUSR2)
(gdb) handle SIGUSR2
Signal Stop Print Pass to program Description
SIGUSR2 Yes Yes Yes User defined signal 2
(gdb) c
Continuing.
Catchpoint 1 (signal SIGUSR2), 0x00007f53174d79b2 in read () from /lib64/libpthread.so.0
(gdb) bt
#0 0x00007f53174d79b2 in read () from /lib64/libpthread.so.0
#1 0x0000000013a74cb4 in read ()
#2 0x0000000013a70950 in snttread ()
#3 0x00000000065a25f5 in nttrd ()
#4 0x00000000064afe40 in nsprecv ()
#5 0x00000000064b8632 in nsrdr ()
#6 0x0000000012bcf1d5 in nsfull_pkt_rcv ()
#7 0x0000000012bd688e in nsfull_brc ()
#8 0x0000000013a5f886 in nioqrc ()
#9 0x00000000135c893a in opikndf2 ()
#10 0x00000000029739c2 in opitsk ()
#11 0x0000000002979ccb in opiino ()
#12 0x00000000135cb7a0 in opiodr ()
#13 0x000000000296fa86 in opidrv ()
#14 0x00000000035da375 in sou2o ()
#15 0x0000000000e0b566 in opimai_real ()
#16 0x00000000035e7011 in ssthrdmain ()
#17 0x0000000000e0b390 in main ()
which is pending to continuation, hence PID 59 waiting for completion.
select s.last_call_et, s.program, event, s.sid, s.serial#, pid, spid
from v$session s, v$process p
where s.paddr=p.addr and lower(s.program) like '%sqlplus%' order by 1;
LAST_CALL_ET PROGRAM EVENT SID SERIAL# PID SPID
------------ ------------------------------ ---------------------------- ------ --------- ---- ------
191 sqlplus.exe SQL*Net message from client 19 50486 36 613512
775 sqlplus@test_db3 (TNS V1-V3) oradebug request completion 829 46358 59 623458
4. strace on DBW0
With strace, we can watch asynchronous I/O (io_submit and io_getevents):
$> strace -t -p 604093
11:29:40 io_submit(140598410809344, 5,
[{data=0x7fdf9f22a738, pwrite, fildes=261, str="\2\242\0\0Z\2\300\0\202\377\216\0355\f\1\4\334\225\0\0\f\0\33\0| \7\0\35E\16\16"..., nbytes=40960, offset=4931584},
{data=0x7fdf9ef17108, pwrite, fildes=261, str="\2\242\0\0qt\301\0\210\377\216\0355\f\2\4I\335\0\0\n\0!\0M\306\n\0`\343\2\2"..., nbytes=40960, offset=781066240},
{data=0x7fdf9eed1758, pwrite, fildes=261, str="\2\242\0\0)\340\323\0J\377\216\0355\f\6\4^\\\0\0\4\0\35\0\210n\10\0\225\210\10\10"..., nbytes=8192, offset=10670645248},
{data=0x7fdf9ee2ae38, pwrite, fildes=261, str="\2\242\0\0I\273\336\0k\377\216\0355\f\1\4\254|\0\0\t\0\27\0\37\373\7\0\272\204\2\2"..., nbytes=8192, offset=16498892800},
{data=0x7fdf9eea4658, pwrite, fildes=2865, str="\6\242\0\0\221\0@\214\227\377\216\0355\f\2\6o$\0\0\1\0\0\0\236\234Z\0\226\377\216\35"..., nbytes=32768, offset=1187840}]) = 5
11:29:40 io_getevents(140598410809344, 1, 128,
[{data=0x7fdf9ee2ae38, obj=0x7fdf9ee2ae38, res=8192, res2=0},
{data=0x7fdf9eed1758, obj=0x7fdf9eed1758, res=8192, res2=0},
{data=0x7fdf9ef17108, obj=0x7fdf9ef17108, res=40960, res2=0},
{data=0x7fdf9f22a738, obj=0x7fdf9f22a738, res=40960, res2=0}],
{600, 0}) = 4
11:29:40 io_getevents(140598410809344, 1, 128,
[{data=0x7fdf9eea4658, obj=0x7fdf9eea4658, res=32768, res2=0}],
{600, 0}) = 1
$ > ./lsof -p 604093
ora_dbw0_ 604093 oracle 261uW REG 8,1 34359730176 688992556 /oratest_db3/oradata/test_db3/undo01.dbf
ora_dbw0_ 604093 oracle 2865uW REG 253,2 5251072 12 /home/oracle/suk/slow_dir3/test_tbs_slow3.dbf
and calls twice io_getevents to read asynchronous I/O events.
(io_submit and io_getevents are matched by "data" and "res" fields. Their order can be different due to asynchronous)
With asynchronous I/O, DBWR calls io_submit to queue I/O requests, kworkers handle actual I/O requests, once I/O operation finishes,
DBWR is signalled and read asynchronous I/O events from the completion queue by io_getevents.
Hence DBWR seems idle even there are heavy I/O activities.
Often "oradebug request completion" is cause by unusually high storage latency or hardware issues, indicating a performance bottleneck.
In AWR reports, we can see those outliers (Av Rd/ Av Writes) of "db file async I/O submit", "db file parallel write" and IOStat sections
(by function, filetype, tablespace,file).
If Linux AIO-MAX-NR is set too (AIO-MAX-NR=131072 AIO-NR=13104), following WARNING wrote into DB Writer trc and DB Performance deteriorated sluggishly. (If currently active AIO number: aio-nr reaches aio-max-nr then io_setup will fail with EAGAIN):
WARNING:1 Oracle process running out of OS kernelI/O resources request_aiolimit=4096 set_aiolimit=0
WARNING:asynch I/O kernel limits is set at AIO-MAX-NR=131072 AIO-NR=13104
select function_name, file_name, disk1_name, count(*) cnt, sum(io_latency) io_latency_SUM, round(avg(io_latency)) io_latency_AVG
from v$io_outlier where disk1_name = 'dm-3' group by function_name, file_name, disk1_name order by file_name, disk1_name;
FUNCTION_NAME FILE_NAME DISK1_NAME CNT IO_LATENCY_SUM IO_LATENCY_AVG
-------------------- -------------------------------------------------- ---------- ---------- -------------- --------------
Buffer Cache Reads /home/oracle/suk/slow_dir3/test_tbs_slow3.dbf dm-3 156 816345 5233
DBWR /home/oracle/suk/slow_dir3/test_tbs_slow3.dbf dm-3 94 1127662 11996
Direct Reads /home/oracle/suk/slow_dir3/test_tbs_slow3.dbf dm-3 4 25645 6411
Direct Writes /home/oracle/suk/slow_dir3/test_tbs_slow3.dbf dm-3 1 10239 10239
Others /home/oracle/suk/slow_dir3/test_tbs_slow3.dbf dm-3 4 21308 5327
$> perf record -e 'workqueue:*' -ag -T
$> perf report
-- Type "Enter" on "19K workqueue:workqueue_queue_work", type "/ Filter symbol by name", then give "io_submit_one"
Samples: 19K of event 'workqueue:workqueue_queue_work', Event count (approx.): 19945
Children Self Command Shared Object Symbol
- 5.85% 0.00% ora_dbw0_test_d [kernel.kallsyms] [k] io_submit_one
- io_submit_one
- 5.77% aio_write
- 5.64% xfs_file_write_iter
- 5.63% xfs_file_dio_write_aligned
- 5.37% iomap_dio_rw
- __iomap_dio_rw
- 5.07% iomap_apply
- 4.91% iomap_dio_bio_actor
- 4.37% iomap_dio_submit_bio.isra.23
- 4.33% submit_bio
- generic_make_request
- 4.16% blk_mq_make_request
- 2.76% blk_flush_plug_list
- blk_mq_flush_plug_list
- 2.76% blk_mq_sched_insert_requests
- 2.70% __blk_mq_delay_run_hw_queue
- 2.70% __blk_mq_run_hw_queue
blk_mq_sched_dispatch_requests
__blk_mq_sched_dispatch_requests
- __blk_mq_do_dispatch_sched
- 2.57% blk_mq_dispatch_rq_list
- 2.56% scsi_queue_rq
- 2.32% pvscsi_queue
_raw_spin_unlock_irqrestore
ret_from_intr
do_IRQ
irq_exit
irq_exit_rcu
__softirqentry_text_start
blk_complete_reqs
scsi_io_completion
- scsi_end_request
- 2.31% blk_update_request
iomap_dio_bio_end_io
queue_work_on
__queue_work
__queue_work
- 1.11% __blk_mq_alloc_request
- 1.11% blk_mq_get_tag
- 1.07% io_schedule
io_schedule_prepare
blk_flush_plug_list
blk_mq_flush_plug_list
blk_mq_sched_insert_requests
kblockd_mod_delayed_work_on
- mod_delayed_work_on
- 1.07% __queue_work
__queue_work
bpftrace -e 'tracepoint:syscalls:sys_enter_io_submit /pid==604093/ {
$iocb_sizeof = sizeof(struct iocb);
$reqs = args->nr;
$iocbpp_ptr = (struct iocb **)args->iocbpp;
time("%H:%M:%S --- ");
printf("%s, iocb_sizeof: %d, io_submit I/O requests: %d\n", comm, $iocb_sizeof, $reqs);
$i = 1; while ($i <= $reqs) { $iocbpp_ptr_2 = (struct iocb **)($iocbpp_ptr + $iocb_sizeof*($i-1));
$iocbpp_struct = (struct iocb *)(*$iocbpp_ptr_2);
printf("--- File: %d --- ptr_2 %p, aio_fildes: %d, aio_nbytes: %d, aio_offset: %ld, aio_lio_opcode(1=PWRITE): %d, aio_data:0X%X\n",
$i, $iocbpp_ptr_2, $iocbpp_struct->aio_fildes, $iocbpp_struct->aio_nbytes, $iocbpp_struct->aio_offset, $iocbpp_struct->aio_lio_opcode, $iocbpp_struct->aio_data);
@All_CNT = count(); @CNT[$iocbpp_struct->aio_fildes] = count(); @SUM[$iocbpp_struct->aio_fildes] = sum( $iocbpp_struct->aio_nbytes);
if ($i > 50) { break; }
$i++ }
printf("\n");}
tracepoint:syscalls:sys_exit_io_getevents /pid==604093/ {time("%H:%M:%S --- "); printf("-- 1 -- io_getevents Ret: %d\n", args->ret); @Ret_CNT = sum(args->ret);
}'
5. ORA-32515 - ORA-32519
Above DB alert.log shows ORA-32515. Here some ORADEBUG related errors:
oracle@test_db3:test_db3 > oerr ora 32515
32515, 00000, "cannot issue ORADEBUG command '%s' to process '%s'; prior command execution time exceeds %s ms"
// *Cause: The process targeted to execute the ORADEBUG command was
// busy executing another ORADEBUG command for a time greater
// than the timeout value specified.
// *Action: Increase the timeout value.
oracle@test_db3:test_db3 > oerr ora 32516
32516, 00000, "cannot wait for process '%s' to finish executing ORADEBUG command '%s'; wait time exceeds %s ms"
// *Cause: The execution of the ORADEBUG command took longer than the
// the timeout value specified.
// *Action: Increase the timeout value.
oracle@test_db3:test_db3 > oerr ora 32517
32517, 00000, "cannot issue ORADEBUG command '%s' to process '%s' (prior command execution time exceeds %s ms); total wait time exceeds %s ms"
// *Cause: The process targeted to execute the ORADEBUG command was
// busy executing another ORADEBUG command. But, because
// the total wait time for all targeted processes
// exceeded the maximum wait time, the ORADEBUG command was
// not issued to the target process.
// *Action: Increase the timeout value.
oracle@test_db3:test_db3 > oerr ora 32518
32518, 00000, "cannot wait for process '%s' to finish executing ORADEBUG command '%s' (waited %s ms); total wait time exceeds %s ms"
// *Cause: The total wait time for all targeted processes exceeded the
// maximum wait time, therefore the wait for the targeted process
// to finish executing the ORADEBUG command was aborted.
// *Action: Increase the timeout value.
oracle@test_db3:test_db3 > oerr ora 32519
32519, 00000, "insufficient privileges to execute ORADEBUG command: %s"
// *Cause: Sufficient privileges were not granted.
// *Action: Perform the operation using a database session that
// has the required privileges. See the Oracle ORADEBUG
// documentation for more details.
Oracle 19.19 Patch:
Patch 25293381: RUNNING "ORADEBUG SETTRACEFILEID TEST" CAUSES ORA-32522: [KGHALO BAD HEAP DS]
And we also noticed that from Oracle 19.26, DBW0 generates a trc file of more than 1 MB.
