SERVER-29606 Add a 'skip' failpoint option

This option allows the failpoint's effect to be skipped for up to a
certain amount of checks.

2 files changed, 36 insertions, 12 deletions

diff --git a/src/mongo/util/fail_point.cpp b/src/mongo/util/fail_point.cpp
index 8f85ba18f17..75be5f48231 100644
--- a/src/mongo/util/fail_point.cpp
+++ b/src/mongo/util/fail_point.cpp
@@ -76,7 +76,7 @@ void FailPoint::setThreadPRNGSeed(int32_t seed) {
     FailPointPRNG::current()->resetSeed(seed);
 }
 
-FailPoint::FailPoint() : _fpInfo(0), _mode(off), _timesOrPeriod(0) {}
+FailPoint::FailPoint() = default;
 
 void FailPoint::shouldFailCloseBlock() {
     _fpInfo.subtractAndFetch(1);
@@ -151,24 +151,28 @@ FailPoint::RetCode FailPoint::slowShouldFailOpenBlock() {
     switch (_mode) {
         case alwaysOn:
             return slowOn;
-
         case random: {
             const AtomicInt32::WordType maxActivationValue = _timesOrPeriod.load();
-            if (FailPointPRNG::current()->nextPositiveInt32() < maxActivationValue) {
+            if (FailPointPRNG::current()->nextPositiveInt32() < maxActivationValue)
                 return slowOn;
-            }
+
             return slowOff;
         }
         case nTimes: {
-            AtomicInt32::WordType newVal = _timesOrPeriod.subtractAndFetch(1);
-
-            if (newVal <= 0) {
+            if (_timesOrPeriod.subtractAndFetch(1) <= 0)
                 disableFailPoint();
-            }
 
             return slowOn;
         }
+        case skip: {
+            // Ensure that once the skip counter reaches within some delta from 0 we don't continue
+            // decrementing it unboundedly because at some point it will roll over and become
+            // positive again
+            if (_timesOrPeriod.load() <= 0 || _timesOrPeriod.subtractAndFetch(1) < 0)
+                return slowOn;
 
+            return slowOff;
+        }
         default:
             error() << "FailPoint Mode not supported: " << static_cast<int>(_mode);
             fassertFailed(16444);
@@ -212,6 +216,23 @@ StatusWith<std::tuple<FailPoint::Mode, FailPoint::ValType, BSONObj>> FailPoint::
                 return {ErrorCodes::BadValue, "'times' option to 'mode' is too large"};
             }
             val = static_cast<int>(longVal);
+        } else if (modeObj.hasField("skip")) {
+            mode = FailPoint::skip;
+
+            long long longVal;
+            auto status = bsonExtractIntegerField(modeObj, "skip", &longVal);
+            if (!status.isOK()) {
+                return status;
+            }
+
+            if (longVal < 0) {
+                return {ErrorCodes::BadValue, "'skip' option to 'mode' must be positive"};
+            }
+
+            if (longVal > std::numeric_limits<int>::max()) {
+                return {ErrorCodes::BadValue, "'skip' option to 'mode' is too large"};
+            }
+            val = static_cast<int>(longVal);
         } else if (modeObj.hasField("activationProbability")) {
             mode = FailPoint::random;
 
diff --git a/src/mongo/util/fail_point.h b/src/mongo/util/fail_point.h
index 77bb9ab882a..a5fc58d21d5 100644
--- a/src/mongo/util/fail_point.h
+++ b/src/mongo/util/fail_point.h
@@ -69,7 +69,7 @@ class FailPoint {
 
 public:
     typedef AtomicUInt32::WordType ValType;
-    enum Mode { off, alwaysOn, random, nTimes };
+    enum Mode { off, alwaysOn, random, nTimes, skip };
     enum RetCode { fastOff = 0, slowOff, slowOn };
 
     /**
@@ -136,6 +136,9 @@ public:
      *           activate.
      *     - nTimes: the number of times this fail point will be active when
      *         #shouldFail or #shouldFailOpenBlock is called.
+     *     - skip: the number of times this failpoint will be inactive when
+     *         #shouldFail or #shouldFailOpenBlock is called. After this number is reached, the
+     *         failpoint will always be active.
      *
      * @param extra arbitrary BSON object that can be stored to this fail point
      *     that can be referenced afterwards with #getData. Defaults to an empty
@@ -155,11 +158,11 @@ private:
     // Bit layout:
     // 31: tells whether this fail point is active.
     // 0~30: unsigned ref counter for active dynamic instances.
-    AtomicUInt32 _fpInfo;
+    AtomicUInt32 _fpInfo{0};
 
     // Invariant: These should be read only if ACTIVE_BIT of _fpInfo is set.
-    Mode _mode;
-    AtomicInt32 _timesOrPeriod;
+    Mode _mode{off};
+    AtomicInt32 _timesOrPeriod{0};
     BSONObj _data;
 
     // protects _mode, _timesOrPeriod, _data