Improve fault injection to MultiRead (#8937)

Summary:
Several improvements to MultiRead:
1. Fix a bug in stress test which causes false positive when both MultiRead() return and individual read request have failure injected.
2. Add two more types of fault that should be handled: empty read results and checksum mismatch
3. Add a message indicating which type of fault is injected
4. Increase the failure rate

Pull Request resolved: https://github.com/facebook/rocksdb/pull/8937

Reviewed By: anand1976

Differential Revision: D31085930

fbshipit-source-id: 3a04994a3cadebf9a64d25e1fe12b14b7a272fba

db_stress_tool/no_batched_ops_stress.cc

@@ -349,7 +349,9 @@ class NonBatchedOpsStressTest : public StressTest {
         // Grab mutex so multiple thread don't try to print the
         // stack trace at the same time
         MutexLock l(thread->shared->GetMutex());
-        fprintf(stderr, "Didn't get expected error from MultiGet\n");
+        fprintf(stderr, "Didn't get expected error from MultiGet. \n");
+        fprintf(stderr, "num_keys %zu Expected %d errors, seen %d\n", num_keys,
+                error_count, stat_nok);
         fprintf(stderr, "Callstack that injected the fault\n");
         fault_fs_guard->PrintFaultBacktrace();
         std::terminate();

tools/db_crashtest.py

@@ -140,7 +140,7 @@ default_params = {
     "continuous_verification_interval" : 0,
     "max_key_len": 3,
     "key_len_percent_dist": "1,30,69",
-    "read_fault_one_in": lambda: random.choice([0, 1000]),
+    "read_fault_one_in": lambda: random.choice([0, 32, 1000]),
     "open_metadata_write_fault_one_in": lambda: random.choice([0, 0, 8]),
     "open_write_fault_one_in": lambda: random.choice([0, 0, 16]),
     "open_read_fault_one_in": lambda: random.choice([0, 0, 32]),

utilities/fault_injection_fs.cc

@@ -26,6 +26,7 @@
 #include "util/coding.h"
 #include "util/crc32c.h"
 #include "util/random.h"
+#include "util/string_util.h"
 #include "util/xxhash.h"
 
 namespace ROCKSDB_NAMESPACE {
@@ -340,7 +341,7 @@ IOStatus TestFSRandomAccessFile::Read(uint64_t offset, size_t n,
   if (s.ok()) {
     s = fs_->InjectThreadSpecificReadError(
         FaultInjectionTestFS::ErrorOperation::kRead, result, use_direct_io(),
-        scratch);
+        scratch, /*need_count_increase=*/true, /*fault_injected=*/nullptr);
   }
   if (s.ok() && fs_->ShouldInjectRandomReadError()) {
     return IOStatus::IOError("Injected read error");
@@ -355,19 +356,25 @@ IOStatus TestFSRandomAccessFile::MultiRead(FSReadRequest* reqs, size_t num_reqs,
     return fs_->GetError();
   }
   IOStatus s = target_->MultiRead(reqs, num_reqs, options, dbg);
+  bool injected_error = false;
   for (size_t i = 0; i < num_reqs; i++) {
     if (!reqs[i].status.ok()) {
       // Already seeing an error.
       break;
     }
+    bool this_injected_error;
     reqs[i].status = fs_->InjectThreadSpecificReadError(
-        FaultInjectionTestFS::ErrorOperation::kRead, &reqs[i].result,
+        FaultInjectionTestFS::ErrorOperation::kMultiReadSingleReq,
-        use_direct_io(), reqs[i].scratch);
+        &(reqs[i].result), use_direct_io(), reqs[i].scratch,
+        /*need_count_increase=*/true,
+        /*fault_injected=*/&this_injected_error);
+    injected_error |= this_injected_error;
   }
   if (s.ok()) {
     s = fs_->InjectThreadSpecificReadError(
-        FaultInjectionTestFS::ErrorOperation::kRead, nullptr, use_direct_io(),
+        FaultInjectionTestFS::ErrorOperation::kMultiRead, nullptr,
-        nullptr);
+        use_direct_io(), nullptr, /*need_count_increase=*/!injected_error,
+        /*fault_injected=*/nullptr);
   }
   if (s.ok() && fs_->ShouldInjectRandomReadError()) {
     return IOStatus::IOError("Injected read error");
@@ -550,7 +557,9 @@ IOStatus FaultInjectionTestFS::NewRandomAccessFile(
     return IOStatus::IOError("Injected error when open random access file");
   }
   IOStatus io_s = InjectThreadSpecificReadError(ErrorOperation::kOpen, nullptr,
-                                                false, nullptr);
+                                                false, nullptr,
+                                                /*need_count_increase=*/true,
+                                                /*fault_injected=*/nullptr);
   if (io_s.ok()) {
     io_s = target()->NewRandomAccessFile(fname, file_opts, result, dbg);
   }
@@ -759,8 +768,11 @@ void FaultInjectionTestFS::UntrackFile(const std::string& f) {
 }
 
 IOStatus FaultInjectionTestFS::InjectThreadSpecificReadError(
-    ErrorOperation /*op*/, Slice* /*result*/, bool /*direct_io*/,
+    ErrorOperation op, Slice* result, bool direct_io, char* /*scratch*/,
-    char* /*scratch*/) {
+    bool need_count_increase, bool* fault_injected) {
+  bool dummy_bool;
+  bool& ret_fault_injected = fault_injected ? *fault_injected : dummy_bool;
+  ret_fault_injected = false;
   ErrorContext* ctx =
         static_cast<ErrorContext*>(thread_local_error_->Get());
   if (ctx == nullptr || !ctx->enable_error_injection || !ctx->one_in) {
@@ -768,12 +780,47 @@ IOStatus FaultInjectionTestFS::InjectThreadSpecificReadError(
   }
 
   if (ctx->rand.OneIn(ctx->one_in)) {
+    if (ctx->count == 0) {
+      ctx->message = "";
+    }
+    if (need_count_increase) {
       ctx->count++;
+    }
     if (ctx->callstack) {
       free(ctx->callstack);
     }
     ctx->callstack = port::SaveStack(&ctx->frames);
+
+    if (op != ErrorOperation::kMultiReadSingleReq) {
+      // Likely non-per read status code for MultiRead
+      ctx->message += "error; ";
+      ret_fault_injected = true;
       return IOStatus::IOError();
+    } else if (Random::GetTLSInstance()->OneIn(8)) {
+      assert(result);
+      // For a small chance, set the failure to status but turn the
+      // result to be empty, which is supposed to be caught for a check.
+      *result = Slice();
+      ctx->message += "inject empty result; ";
+      ret_fault_injected = true;
+    } else if (!direct_io && Random::GetTLSInstance()->OneIn(7)) {
+      assert(result);
+      // With direct I/O, many extra bytes might be read so corrupting
+      // one byte might not cause checksum mismatch. Skip checksum
+      // corruption injection.
+      // For a small chance, set the failure to status but corrupt the
+      // result in a way that checksum checking is supposed to fail.
+      // Corrupt the last byte, which is supposed to be a checksum byte
+      // It would work for CRC. Not 100% sure for xxhash and will adjust
+      // if it is not the case.
+      const_cast<char*>(result->data())[result->size() - 1]++;
+      ctx->message += "corrupt last byte; ";
+      ret_fault_injected = true;
+    } else {
+      ctx->message += "error result multiget single; ";
+      ret_fault_injected = true;
+      return IOStatus::IOError();
+    }
   }
   return IOStatus::OK();
 }
@@ -835,6 +882,7 @@ void FaultInjectionTestFS::PrintFaultBacktrace() {
     return;
   }
   fprintf(stderr, "Injected error type = %d\n", ctx->type);
+  fprintf(stderr, "Message: %s\n", ctx->message.c_str());
   port::PrintAndFreeStack(ctx->callstack, ctx->frames);
   ctx->callstack = nullptr;
 #endif

utilities/fault_injection_fs.h

@@ -370,6 +370,8 @@ class FaultInjectionTestFS : public FileSystemWrapper {
   // Specify what the operation, so we can inject the right type of error
   enum ErrorOperation : char {
     kRead = 0,
+    kMultiReadSingleReq = 1,
+    kMultiRead = 2,
     kOpen,
   };
 
@@ -440,8 +442,12 @@ class FaultInjectionTestFS : public FileSystemWrapper {
   // corruption in the contents of scratch, or truncation of slice
   // are the types of error with equal probability. For OPEN,
   // its always an IOError.
+  // fault_injected returns whether a fault is injected. It is needed
+  // because some fault is inected with IOStatus to be OK.
   IOStatus InjectThreadSpecificReadError(ErrorOperation op, Slice* slice,
-                                         bool direct_io, char* scratch);
+                                         bool direct_io, char* scratch,
+                                         bool need_count_increase,
+                                         bool* fault_injected);
 
   // Get the count of how many times we injected since the previous call
   int GetAndResetErrorCount() {
@@ -525,6 +531,7 @@ class FaultInjectionTestFS : public FileSystemWrapper {
     int count;
     bool enable_error_injection;
     void* callstack;
+    std::string message;
     int frames;
     ErrorType type;