Fix statsd crash due to bad bucket index
Statsd crashes because predictAnomalyTimestamp requests past buckets
that occurred before time began. That's fine, but statsd needs to know
that the data before time began was 0 (instead of reading before the
beginning of the array and crashing).
AnomalyTracker's use of bucketNumbers is in general risky, and should be
auditted. But this cl will fix the current crashing.
Bug: 73825954
Test: Definitely necessary. Will write during audit.
Change-Id: I990ff134153f290d3089bfe3440d838f47996b63
diff --git a/cmds/statsd/src/anomaly/AnomalyTracker.cpp b/cmds/statsd/src/anomaly/AnomalyTracker.cpp
index 6ee51f3..133f33b 100644
--- a/cmds/statsd/src/anomaly/AnomalyTracker.cpp
+++ b/cmds/statsd/src/anomaly/AnomalyTracker.cpp
@@ -59,6 +59,12 @@
}
size_t AnomalyTracker::index(int64_t bucketNum) const {
+ if (bucketNum < 0) {
+ // To support this use-case, we can easily modify index to wrap around. But currently
+ // AnomalyTracker should never need this, so if it happens, it's a bug we should log.
+ // TODO: Audit this.
+ ALOGE("index() was passed a negative bucket number (%lld)!", (long long)bucketNum);
+ }
return bucketNum % mNumOfPastBuckets;
}
@@ -72,9 +78,7 @@
// The past packets are ancient. Empty out old mPastBuckets[i] values and reset
// mSumOverPastBuckets.
if (latestPastBucketNum - mMostRecentBucketNum >= mNumOfPastBuckets) {
- mPastBuckets.clear();
- mPastBuckets.resize(mNumOfPastBuckets);
- mSumOverPastBuckets.clear();
+ resetStorage();
} else {
for (int64_t i = std::max(0LL, (long long)(mMostRecentBucketNum - mNumOfPastBuckets + 1));
i <= latestPastBucketNum - mNumOfPastBuckets; i++) {
@@ -150,7 +154,7 @@
int64_t AnomalyTracker::getPastBucketValue(const MetricDimensionKey& key,
const int64_t& bucketNum) const {
- if (mNumOfPastBuckets == 0) {
+ if (mNumOfPastBuckets == 0 || bucketNum < 0) {
return 0;
}