SERVER-15111 Treat corruption of final journal section as an expected event

2 files changed, 89 insertions, 94 deletions

diff --git a/jstests/dur/checksum.js b/jstests/dur/checksum.js
new file mode 100644
index 00000000000..1270e337a87
--- /dev/null
+++ b/jstests/dur/checksum.js
@@ -0,0 +1,89 @@
+// Test checksum validation of journal files.
+
+var testname = "dur_checksum";
+var path = MongoRunner.dataPath + testname;
+
+if (0) {
+    // This is used to create the prototype journal file.
+    jsTest.log("Just creating prototype journal, not testing anything");
+    var conn = startMongodEmpty("--port", 30001, "--dbpath", path, "--dur");
+    var db = conn.getDB("test");
+
+    // each insert is in it's own commit.
+    db.foo.insert({a: 1});
+    db.runCommand({getlasterror:1, j:1})
+
+    db.foo.insert({a: 2});
+    db.runCommand({getlasterror:1, j:1})
+
+    stopMongod(30001, /*signal*/9);
+
+    jsTest.log("Journal file left at " + path + "/journal/j._0");
+    quit();
+    // A hex editor must be used to replace the checksums of specific journal sections with
+    // "0BADC0DE 1BADC0DE 2BADC0DE 3BADCD0E"
+}
+
+function startMongodWithJournal() {
+    return startMongodNoReset("--port", 30001,
+                              "--dbpath", path,
+                              "--dur",
+                              "--smallfiles",
+                              "--durOptions", 1 /*DurDumpJournal*/);
+}
+
+
+jsTest.log("Starting with good.journal to make sure everything works");
+resetDbpath(path);
+mkdir(path + '/journal');
+copyFile("jstests/libs/dur_checksum_good.journal", path + "/journal/j._0");
+var conn = startMongodWithJournal();
+var db = conn.getDB('test');
+assert.eq(db.foo.count(), 2);
+stopMongod(30001);
+
+
+// dur_checksum_bad_last.journal is good.journal with the bad checksum on the last section.
+jsTest.log("Starting with bad_last.journal");
+resetDbpath(path);
+mkdir(path + '/journal');
+copyFile("jstests/libs/dur_checksum_bad_last.journal", path + "/journal/j._0");
+conn = startMongodWithJournal();
+var db = conn.getDB('test');
+assert.eq(db.foo.count(), 1); // 2nd insert "never happened"
+stopMongod(30001);
+
+
+// dur_checksum_bad_first.journal is good.journal with the bad checksum on the prior section.
+// This means there is a good commit after the bad one. We currently ignore this, but a future
+// version of the server may be able to detect this case.
+jsTest.log("Starting with bad_first.journal");
+resetDbpath(path);
+mkdir(path + '/journal');
+copyFile("jstests/libs/dur_checksum_bad_first.journal", path + "/journal/j._0");
+conn = startMongodWithJournal();
+var db = conn.getDB('test');
+assert.eq(db.foo.count(), 0); // Neither insert happened.
+stopMongod(30001);
+
+// If we detect an error in a non-final journal file, that is considered an error.
+jsTest.log("Starting with bad_last.journal followed by good.journal");
+resetDbpath(path);
+mkdir(path + '/journal');
+copyFile("jstests/libs/dur_checksum_bad_first.journal", path + "/journal/j._0");
+copyFile("jstests/libs/dur_checksum_good.journal", path + "/journal/j._1");
+
+exitCode = runMongoProgram("mongod",
+                           "--port", 30001,
+                           "--dbpath", path,
+                           "--dur",
+                           "--smallfiles",
+                           "--durOptions", 1 /*DurDumpJournal*/
+                                         + 2 /*DurScanOnly*/);
+
+assert.eq(exitCode, 100 /*EXIT_UNCAUGHT*/);
+
+// TODO Possibly we could check the mongod log to verify that the correct type of exception was
+// thrown.  But that would introduce a dependency on the mongod log format, which we may not want.
+
+jsTest.log("SUCCESS checksum.js");
diff --git a/jstests/dur/md5.js b/jstests/dur/md5.js
deleted file mode 100644
index 1b4ec43340e..00000000000
--- a/jstests/dur/md5.js
+++ /dev/null
@@ -1,94 +0,0 @@
-/**
- * Test md5 validation of journal file.
- * This test is dependent on the journal file format and may require an update if the format changes,
- * see comments near fuzzFile() below.
- */
-
-var debugging = false;
-var testname = "dur_md5";
-var step = 1;
-var conn = null;
-
-function log(str) {
-    print();
-    if(str)
-        print(testname+" step " + step++ + " " + str);
-    else
-        print(testname+" step " + step++);
-}
-
-/** Changes here may require updating the byte index of the md5 hash, see File comments below. */
-function work() {
-    log("work");
-    var d = conn.getDB("test");
-    d.foo.insert({ _id: 3, x: 22 });
-    d.foo.insert({ _id: 4, x: 22 });
-    d.a.insert({ _id: 3, x: 22, y: [1, 2, 3] });
-    d.a.insert({ _id: 4, x: 22, y: [1, 2, 3] });
-    d.a.update({ _id: 4 }, { $inc: { x: 1} });
-    
-    // try building an index.  however, be careful as object id's in system.indexes would vary, so we do it manually:
-    d.system.indexes.insert({ _id: 99, ns: "test.a", key: { x: 1 }, name: "x_1", v: 0 });
-    log("endwork");
-}
-
-if( debugging ) { 
-    // mongod already running in debugger
-    conn = db.getMongo();
-    work();
-    sleep(30000);
-    quit();
-}
-
-log();
-
-var path = MongoRunner.dataPath + testname+"dur";
-
-log();
-conn = startMongodEmpty("--port", 30001, "--dbpath", path, "--dur", "--smallfiles", "--durOptions", 8);
-work();
-
-// wait for group commit.
-printjson(conn.getDB('admin').runCommand({getlasterror:1, fsync:1}));

-

-log("kill -9");
-
-// kill the process hard
-stopMongod(30001, /*signal*/9);
-
-// journal file should be present, and non-empty as we killed hard
-
-// Bit flip the first byte of the md5sum contained within the opcode footer.

-// This ensures we get an md5 exception instead of some other type of exception.

-var file = path + "/journal/j._0";

-

-// if test fails, uncomment these "cp" lines to debug:

-// run("cp", file, "/tmp/before");

-

-// journal header is 8192

-// jsectheader is 20

-// so a little beyond that

-fuzzFile(file, 8214+8);

-

-// run("cp", file, "/tmp/after");
-
-log("run mongod again recovery should fail");
-
-// 100 exit code corresponds to EXIT_UNCAUGHT, which is triggered when there is an exception during recovery.
-// 14 is is sometimes triggered instead due to SERVER-2184
-exitCode = runMongoProgram( "mongod", "--port", 30002, "--dbpath", path, "--dur", "--smallfiles", "--durOptions", /*9*/13 );

-

-if (exitCode != 100 && exitCode != 14) {

-    print("\n\n\nFAIL md5.js expected mongod to fail but didn't? mongod exitCode: " + exitCode + "\n\n\n");

-    // sleep a little longer to get more output maybe

-    sleep(2000);

-    assert(false);

-}
-
-// TODO Possibly we could check the mongod log to verify that the correct type of exception was thrown.  But
-// that would introduce a dependency on the mongod log format, which we may not want.

-

-print("SUCCESS md5.js");

-

-// if we sleep a littler here we may get more out the mongod output logged

-sleep(500);