runtime/pprof: avoid crash due to truncated stack traces
The profile proto message builder maintains a location entry cache
that maps a location (possibly involving multiple user frames
that represent inlined function calls) to the location id. We have
been using the first pc of the inlined call sequence as the key of
the cached location entry assuming that, for a given pc, the sequence
of frames representing the inlined call stack is deterministic and
stable. Then, when analyzing the new stack trace, we expected the
exact number of pcs to be present in the captured stack trace upon
the cache hit.
This assumption does not hold, however, in the presence of the stack
trace truncation in the runtime during profiling, and also with the
potential bugs in runtime.
A better fix is to use all the pcs of the inlined call sequece as
the key instead of the first pc. But that is a bigger code change.
This CL avoids the crash assuming the trace was truncated.
Fixes #35538
Change-Id: I8c6bae98bc8b178ee51523c7316f56b1cce6df16
Reviewed-on: https://go-review.googlesource.com/c/go/+/207609
Run-TryBot: Hyang-Ah Hana Kim <hyangah@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
diff --git a/src/runtime/pprof/pprof_test.go b/src/runtime/pprof/pprof_test.go
index 64e03ae..20b44e1 100644
--- a/src/runtime/pprof/pprof_test.go
+++ b/src/runtime/pprof/pprof_test.go
@@ -1183,6 +1183,32 @@
{Value: []int64{30, 30 * period}, Location: []*profile.Location{{ID: 1}, {ID: 1}}},
{Value: []int64{40, 40 * period}, Location: []*profile.Location{{ID: 1}}},
},
+ }, {
+ name: "truncated_stack_trace_later",
+ input: []uint64{
+ 3, 0, 500, // hz = 500. Must match the period.
+ 5, 0, 50, inlinedCalleePtr, inlinedCallerPtr,
+ 4, 0, 60, inlinedCalleePtr,
+ },
+ wantLocs: [][]string{{"runtime/pprof.inlinedCallee", "runtime/pprof.inlinedCaller"}},
+ wantSamples: []*profile.Sample{
+ {Value: []int64{50, 50 * period}, Location: []*profile.Location{{ID: 1}}},
+ {Value: []int64{60, 60 * period}, Location: []*profile.Location{{ID: 1}}},
+ },
+ }, {
+ name: "truncated_stack_trace_first",
+ input: []uint64{
+ 3, 0, 500, // hz = 500. Must match the period.
+ 4, 0, 70, inlinedCalleePtr,
+ 5, 0, 80, inlinedCalleePtr, inlinedCallerPtr,
+ },
+ wantLocs: [][]string{ // the inline info is screwed up, but better than a crash.
+ {"runtime/pprof.inlinedCallee"},
+ {"runtime/pprof.inlinedCaller"}},
+ wantSamples: []*profile.Sample{
+ {Value: []int64{70, 70 * period}, Location: []*profile.Location{{ID: 1}}},
+ {Value: []int64{80, 80 * period}, Location: []*profile.Location{{ID: 1}, {ID: 2}}},
+ },
}}
for _, tc := range testCases {
diff --git a/src/runtime/pprof/proto.go b/src/runtime/pprof/proto.go
index a42cd80..8a30c71 100644
--- a/src/runtime/pprof/proto.go
+++ b/src/runtime/pprof/proto.go
@@ -394,7 +394,23 @@
// then, record the cached location.
locs = append(locs, l.id)
- stk = stk[len(l.pcs):] // skip the matching pcs.
+
+ // The stk may be truncated due to the stack depth limit
+ // (e.g. See maxStack and maxCPUProfStack in runtime) or
+ // bugs in runtime. Avoid the crash in either case.
+ // TODO(hyangah): The correct fix may require using the exact
+ // pcs as the key for b.locs cache management instead of just
+ // relying on the very first pc. We are late in the go1.14 dev
+ // cycle, so this is a workaround with little code change.
+ if len(l.pcs) > len(stk) {
+ stk = nil
+ // TODO(hyangah): would be nice if we can enable
+ // debug print out on demand and report the problematic
+ // cached location entry and stack traces. Do we already
+ // have such facility to utilize (e.g. GODEBUG)?
+ } else {
+ stk = stk[len(l.pcs):] // skip the matching pcs.
+ }
continue
}
