[
https://issues.apache.org/jira/browse/IMPALA-8264?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16781947#comment-16781947
]
ASF subversion and git services commented on IMPALA-8264:
---------------------------------------------------------
Commit 74dba7a52da6dc99d450e3a7d789ceac3197bdfb in impala's branch
refs/heads/master from Lars Volker
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=74dba7a ]
IMPALA-8264: Fix overflow in SystemStateInfo::ComputeCpuRatios
We pass '0' as the initial value to std::accumulate(), which causes it
to use int32_t for its internal state, which then overflows when values
read from /proc/stat get too large. The fix is to initialize it with
'0L' to make it use int64_t.
Change-Id: I46dfb7264e85a3da492e383caaa6853d8ef95f35
Reviewed-on: http://gerrit.cloudera.org:8080/12638
Reviewed-by: Impala Public Jenkins <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>
> system-state-info.cc:102] Check failed: total_tics > 0 (-4294962910 vs. 0)
> --------------------------------------------------------------------------
>
> Key: IMPALA-8264
> URL: https://issues.apache.org/jira/browse/IMPALA-8264
> Project: IMPALA
> Issue Type: Bug
> Components: Backend
> Affects Versions: Impala 3.2.0
> Reporter: Michael Ho
> Assignee: Lars Volker
> Priority: Critical
>
> When running 64-streams concurrent TPC-DS in debug builds at commit
> 681b256627e5e569778d4996bbdf530e30c528eb in a 140 node cluster, hit the
> following DCHECK after close to 14 hours of running.
> {noformat}
> F0228 00:42:35.132514 139036 system-state-info.cc:102] Check failed:
> total_tics > 0 (-4294962910 vs. 0)
> {noformat}
> {noformat}
> void SystemStateInfo::ComputeCpuRatios() {
> const CpuValues& cur = cpu_values_[cur_val_idx_];
> const CpuValues& old = cpu_values_[1 - cur_val_idx_];
> // Sum up all counters
> int64_t cur_sum = accumulate(cur.begin(), cur.end(), 0);
> int64_t old_sum = accumulate(old.begin(), old.end(), 0);
> int64_t total_tics = cur_sum - old_sum;
> // If less than 1/USER_HZ time has time has passed for any of the counters,
> the ratio is
> // zero (to avoid dividing by zero).
> if (total_tics == 0) {
> memset(&cpu_ratios_, 0, sizeof(cpu_ratios_));
> return;
> }
> DCHECK_GT(total_tics, 0); <<----
> {noformat}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
