Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commit9f75e37

Browse files
committed
Refactor code to print pgbench progress reports.
threadRun() function is very long and deeply-nested. Extract the code toprint progress reports to a separate function, to make it slightly easierto read.Author: Fabien CoelhoDiscussion:https://www.postgresql.org/message-id/alpine.DEB.2.21.1903101225270.17271%40lancre
1 parent5857be9 commit9f75e37

File tree

1 file changed

+94
-83
lines changed

1 file changed

+94
-83
lines changed

‎src/bin/pgbench/pgbench.c

Lines changed: 94 additions & 83 deletions
Original file line numberDiff line numberDiff line change
@@ -4893,6 +4893,99 @@ addScript(ParsedScript script)
48934893
num_scripts++;
48944894
}
48954895

4896+
/*
4897+
* Print progress report.
4898+
*
4899+
* On entry, *last and *last_report contain the statistics and time of last
4900+
* progress report. On exit, they are updated with the new stats.
4901+
*/
4902+
staticvoid
4903+
printProgressReport(TState*thread,int64test_start,int64now,
4904+
StatsData*last,int64*last_report)
4905+
{
4906+
/* generate and show report */
4907+
int64run=now-*last_report,
4908+
ntx;
4909+
doubletps,
4910+
total_run,
4911+
latency,
4912+
sqlat,
4913+
lag,
4914+
stdev;
4915+
chartbuf[315];
4916+
StatsDatacur;
4917+
4918+
/*
4919+
* Add up the statistics of all threads.
4920+
*
4921+
* XXX: No locking. There is no guarantee that we get an atomic snapshot
4922+
* of the transaction count and latencies, so these figures can well be
4923+
* off by a small amount. The progress report's purpose is to give a
4924+
* quick overview of how the test is going, so that shouldn't matter too
4925+
* much. (If a read from a 64-bit integer is not atomic, you might get a
4926+
* "torn" read and completely bogus latencies though!)
4927+
*/
4928+
initStats(&cur,0);
4929+
for (inti=0;i<nthreads;i++)
4930+
{
4931+
mergeSimpleStats(&cur.latency,&thread[i].stats.latency);
4932+
mergeSimpleStats(&cur.lag,&thread[i].stats.lag);
4933+
cur.cnt+=thread[i].stats.cnt;
4934+
cur.skipped+=thread[i].stats.skipped;
4935+
}
4936+
4937+
/* we count only actually executed transactions */
4938+
ntx= (cur.cnt-cur.skipped)- (last->cnt-last->skipped);
4939+
total_run= (now-test_start) /1000000.0;
4940+
tps=1000000.0*ntx /run;
4941+
if (ntx>0)
4942+
{
4943+
latency=0.001* (cur.latency.sum-last->latency.sum) /ntx;
4944+
sqlat=1.0* (cur.latency.sum2-last->latency.sum2) /ntx;
4945+
stdev=0.001*sqrt(sqlat-1000000.0*latency*latency);
4946+
lag=0.001* (cur.lag.sum-last->lag.sum) /ntx;
4947+
}
4948+
else
4949+
{
4950+
latency=sqlat=stdev=lag=0;
4951+
}
4952+
4953+
if (progress_timestamp)
4954+
{
4955+
/*
4956+
* On some platforms the current system timestamp is available in
4957+
* now_time, but rather than get entangled with that, we just eat the
4958+
* cost of an extra syscall in all cases.
4959+
*/
4960+
structtimevaltv;
4961+
4962+
gettimeofday(&tv,NULL);
4963+
snprintf(tbuf,sizeof(tbuf),"%ld.%03ld s",
4964+
(long)tv.tv_sec, (long) (tv.tv_usec /1000));
4965+
}
4966+
else
4967+
{
4968+
/* round seconds are expected, but the thread may be late */
4969+
snprintf(tbuf,sizeof(tbuf),"%.1f s",total_run);
4970+
}
4971+
4972+
fprintf(stderr,
4973+
"progress: %s, %.1f tps, lat %.3f ms stddev %.3f",
4974+
tbuf,tps,latency,stdev);
4975+
4976+
if (throttle_delay)
4977+
{
4978+
fprintf(stderr,", lag %.3f ms",lag);
4979+
if (latency_limit)
4980+
fprintf(stderr,", "INT64_FORMAT" skipped",
4981+
cur.skipped-last->skipped);
4982+
}
4983+
fprintf(stderr,"\n");
4984+
4985+
*last=cur;
4986+
*last_report=now;
4987+
}
4988+
48964989
staticvoid
48974990
printSimpleStats(constchar*prefix,SimpleStats*ss)
48984991
{
@@ -6199,89 +6292,7 @@ threadRun(void *arg)
61996292
now=INSTR_TIME_GET_MICROSEC(now_time);
62006293
if (now >=next_report)
62016294
{
6202-
/* generate and show report */
6203-
StatsDatacur;
6204-
int64run=now-last_report,
6205-
ntx;
6206-
doubletps,
6207-
total_run,
6208-
latency,
6209-
sqlat,
6210-
lag,
6211-
stdev;
6212-
chartbuf[315];
6213-
6214-
/*
6215-
* Add up the statistics of all threads.
6216-
*
6217-
* XXX: No locking. There is no guarantee that we get an
6218-
* atomic snapshot of the transaction count and latencies, so
6219-
* these figures can well be off by a small amount. The
6220-
* progress report's purpose is to give a quick overview of
6221-
* how the test is going, so that shouldn't matter too much.
6222-
* (If a read from a 64-bit integer is not atomic, you might
6223-
* get a "torn" read and completely bogus latencies though!)
6224-
*/
6225-
initStats(&cur,0);
6226-
for (i=0;i<nthreads;i++)
6227-
{
6228-
mergeSimpleStats(&cur.latency,&thread[i].stats.latency);
6229-
mergeSimpleStats(&cur.lag,&thread[i].stats.lag);
6230-
cur.cnt+=thread[i].stats.cnt;
6231-
cur.skipped+=thread[i].stats.skipped;
6232-
}
6233-
6234-
/* we count only actually executed transactions */
6235-
ntx= (cur.cnt-cur.skipped)- (last.cnt-last.skipped);
6236-
total_run= (now-thread_start) /1000000.0;
6237-
tps=1000000.0*ntx /run;
6238-
if (ntx>0)
6239-
{
6240-
latency=0.001* (cur.latency.sum-last.latency.sum) /ntx;
6241-
sqlat=1.0* (cur.latency.sum2-last.latency.sum2) /ntx;
6242-
stdev=0.001*sqrt(sqlat-1000000.0*latency*latency);
6243-
lag=0.001* (cur.lag.sum-last.lag.sum) /ntx;
6244-
}
6245-
else
6246-
{
6247-
latency=sqlat=stdev=lag=0;
6248-
}
6249-
6250-
if (progress_timestamp)
6251-
{
6252-
/*
6253-
* On some platforms the current system timestamp is
6254-
* available in now_time, but rather than get entangled
6255-
* with that, we just eat the cost of an extra syscall in
6256-
* all cases.
6257-
*/
6258-
structtimevaltv;
6259-
6260-
gettimeofday(&tv,NULL);
6261-
snprintf(tbuf,sizeof(tbuf),"%ld.%03ld s",
6262-
(long)tv.tv_sec, (long) (tv.tv_usec /1000));
6263-
}
6264-
else
6265-
{
6266-
/* round seconds are expected, but the thread may be late */
6267-
snprintf(tbuf,sizeof(tbuf),"%.1f s",total_run);
6268-
}
6269-
6270-
fprintf(stderr,
6271-
"progress: %s, %.1f tps, lat %.3f ms stddev %.3f",
6272-
tbuf,tps,latency,stdev);
6273-
6274-
if (throttle_delay)
6275-
{
6276-
fprintf(stderr,", lag %.3f ms",lag);
6277-
if (latency_limit)
6278-
fprintf(stderr,", "INT64_FORMAT" skipped",
6279-
cur.skipped-last.skipped);
6280-
}
6281-
fprintf(stderr,"\n");
6282-
6283-
last=cur;
6284-
last_report=now;
6295+
printProgressReport(thread,thread_start,now,&last,&last_report);
62856296

62866297
/*
62876298
* Ensure that the next report is in the future, in case

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp