X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/192e5a9d12fb9fd16baff2b61af011611b671983..d542de20d8617f5823ab8f675c114f78aaf4a924:/services/datamanager/datamanager_test.go diff --git a/services/datamanager/datamanager_test.go b/services/datamanager/datamanager_test.go index 26a2fbf92b..7a8fff5c32 100644 --- a/services/datamanager/datamanager_test.go +++ b/services/datamanager/datamanager_test.go @@ -6,17 +6,20 @@ import ( "git.curoverse.com/arvados.git/sdk/go/arvadosclient" "git.curoverse.com/arvados.git/sdk/go/arvadostest" "git.curoverse.com/arvados.git/sdk/go/keepclient" + "git.curoverse.com/arvados.git/services/datamanager/collection" + "git.curoverse.com/arvados.git/services/datamanager/summary" "io/ioutil" "net/http" "os" "os/exec" + "path" "regexp" "strings" "testing" "time" ) -var arv arvadosclient.ArvadosClient +var arv *arvadosclient.ArvadosClient var keepClient *keepclient.KeepClient var keepServers []string @@ -29,7 +32,7 @@ func SetupDataManagerTest(t *testing.T) { arvadostest.StartKeep(2, false) var err error - arv, err = makeArvadosClient() + arv, err = arvadosclient.MakeArvadosClient() if err != nil { t.Fatalf("Error making arvados client: %s", err) } @@ -37,9 +40,8 @@ func SetupDataManagerTest(t *testing.T) { // keep client keepClient = &keepclient.KeepClient{ - Arvados: &arv, + Arvados: arv, Want_replicas: 2, - Using_proxy: true, Client: &http.Client{}, } @@ -56,6 +58,8 @@ func SetupDataManagerTest(t *testing.T) { func TearDownDataManagerTest(t *testing.T) { arvadostest.StopKeep(2) arvadostest.StopAPI() + summary.WriteDataTo = "" + collection.HeapProfileFilename = "" } func putBlock(t *testing.T, data string) string { @@ -254,13 +258,10 @@ func valueInArray(value string, list []string) bool { return false } -/* -Test env uses two keep volumes. The volume names can be found by reading the files - ARVADOS_HOME/tmp/keep0.volume and ARVADOS_HOME/tmp/keep1.volume - -The keep volumes are of the dir structure: - volumeN/subdir/locator -*/ +// Test env uses two keep volumes. The volume names can be found by reading the files +// ARVADOS_HOME/tmp/keep0.volume and ARVADOS_HOME/tmp/keep1.volume +// +// The keep volumes are of the dir structure: volumeN/subdir/locator func backdateBlocks(t *testing.T, oldUnusedBlockLocators []string) { // First get rid of any size hints in the locators var trimmedBlockLocators []string @@ -342,11 +343,9 @@ func waitUntilQueuesFinishWork(t *testing.T) { } } -/* -Create some blocks and backdate some of them. -Also create some collections and delete some of them. -Verify block indexes. -*/ +// Create some blocks and backdate some of them. +// Also create some collections and delete some of them. +// Verify block indexes. func TestPutAndGetBlocks(t *testing.T) { defer TearDownDataManagerTest(t) SetupDataManagerTest(t) @@ -404,6 +403,10 @@ func TestPutAndGetBlocks(t *testing.T) { t.Fatalf("Locators for both these collections expected to be same: %s %s", oneOfTwoWithSameDataLocator, secondOfTwoWithSameDataLocator) } + // create collection with empty manifest text + emptyBlockLocator := putBlock(t, "") + emptyCollection := createCollection(t, "") + // Verify blocks before doing any backdating / deleting. var expected []string expected = append(expected, oldUnusedBlockLocators...) @@ -412,6 +415,7 @@ func TestPutAndGetBlocks(t *testing.T) { expected = append(expected, replicationCollectionLocator) expected = append(expected, oneOfTwoWithSameDataLocator) expected = append(expected, secondOfTwoWithSameDataLocator) + expected = append(expected, emptyBlockLocator) verifyBlocks(t, nil, expected, 2) @@ -425,6 +429,8 @@ func TestPutAndGetBlocks(t *testing.T) { backdateBlocks(t, oldUnusedBlockLocators) deleteCollection(t, toBeDeletedCollectionUUID) deleteCollection(t, secondOfTwoWithSameDataUUID) + backdateBlocks(t, []string{emptyBlockLocator}) + deleteCollection(t, emptyCollection) // Run data manager again dataManagerSingleRun(t) @@ -437,6 +443,7 @@ func TestPutAndGetBlocks(t *testing.T) { expected = append(expected, toBeDeletedCollectionLocator) expected = append(expected, oneOfTwoWithSameDataLocator) expected = append(expected, secondOfTwoWithSameDataLocator) + expected = append(expected, emptyBlockLocator) // even when unreferenced, this remains verifyBlocks(t, oldUnusedBlockLocators, expected, 2) @@ -526,3 +533,200 @@ func TestRunDatamanagerAsNonAdminUser(t *testing.T) { t.Fatalf("Expected error during singlerun as non-admin user") } } + +func TestPutAndGetBlocks_NoErrorDuringSingleRun(t *testing.T) { + testOldBlocksNotDeletedOnDataManagerError(t, "", "", false, false) +} + +func TestPutAndGetBlocks_ErrorDuringGetCollectionsBadWriteTo(t *testing.T) { + badpath, err := arvadostest.CreateBadPath() + if err != nil { + t.Fatalf(err.Error()) + } + defer func() { + err = arvadostest.DestroyBadPath(badpath) + if err != nil { + t.Fatalf(err.Error()) + } + }() + testOldBlocksNotDeletedOnDataManagerError(t, path.Join(badpath, "writetofile"), "", true, true) +} + +func TestPutAndGetBlocks_ErrorDuringGetCollectionsBadHeapProfileFilename(t *testing.T) { + badpath, err := arvadostest.CreateBadPath() + if err != nil { + t.Fatalf(err.Error()) + } + defer func() { + err = arvadostest.DestroyBadPath(badpath) + if err != nil { + t.Fatalf(err.Error()) + } + }() + testOldBlocksNotDeletedOnDataManagerError(t, "", path.Join(badpath, "heapprofilefile"), true, true) +} + +// Create some blocks and backdate some of them. +// Run datamanager while producing an error condition. +// Verify that the blocks are hence not deleted. +func testOldBlocksNotDeletedOnDataManagerError(t *testing.T, writeDataTo string, heapProfileFile string, expectError bool, expectOldBlocks bool) { + defer TearDownDataManagerTest(t) + SetupDataManagerTest(t) + + // Put some blocks and backdate them. + var oldUnusedBlockLocators []string + oldUnusedBlockData := "this block will have older mtime" + for i := 0; i < 5; i++ { + oldUnusedBlockLocators = append(oldUnusedBlockLocators, putBlock(t, fmt.Sprintf("%s%d", oldUnusedBlockData, i))) + } + backdateBlocks(t, oldUnusedBlockLocators) + + // Run data manager + summary.WriteDataTo = writeDataTo + collection.HeapProfileFilename = heapProfileFile + + err := singlerun(arv) + if !expectError { + if err != nil { + t.Fatalf("Got an error during datamanager singlerun: %v", err) + } + } else { + if err == nil { + t.Fatalf("Expected error during datamanager singlerun") + } + } + waitUntilQueuesFinishWork(t) + + // Get block indexes and verify that all backdated blocks are not/deleted as expected + if expectOldBlocks { + verifyBlocks(t, nil, oldUnusedBlockLocators, 2) + } else { + verifyBlocks(t, oldUnusedBlockLocators, nil, 2) + } +} + +// Create a collection with multiple streams and blocks +func createMultiStreamBlockCollection(t *testing.T, data string, numStreams, numBlocks int) (string, []string) { + defer switchToken(arvadostest.AdminToken)() + + manifest := "" + locators := make(map[string]bool) + for s := 0; s < numStreams; s++ { + manifest += fmt.Sprintf("./stream%d ", s) + for b := 0; b < numBlocks; b++ { + locator, _, err := keepClient.PutB([]byte(fmt.Sprintf("%s in stream %d and block %d", data, s, b))) + if err != nil { + t.Fatalf("Error creating block %d in stream %d: %v", b, s, err) + } + locators[strings.Split(locator, "+A")[0]] = true + manifest += locator + " " + } + manifest += "0:1:dummyfile.txt\n" + } + + collection := make(Dict) + err := arv.Create("collections", + arvadosclient.Dict{"collection": arvadosclient.Dict{"manifest_text": manifest}}, + &collection) + + if err != nil { + t.Fatalf("Error creating collection %v", err) + } + + var locs []string + for k := range locators { + locs = append(locs, k) + } + + return collection["uuid"].(string), locs +} + +// Create collection with multiple streams and blocks; backdate the blocks and but do not delete the collection. +// Also, create stray block and backdate it. +// After datamanager run: expect blocks from the collection, but not the stray block. +func TestManifestWithMultipleStreamsAndBlocks(t *testing.T) { + testManifestWithMultipleStreamsAndBlocks(t, 100, 10, "", false) +} + +// Same test as TestManifestWithMultipleStreamsAndBlocks with an additional +// keepstore of a service type other than "disk". Only the "disk" type services +// will be indexed by datamanager and hence should work the same way. +func TestManifestWithMultipleStreamsAndBlocks_WithOneUnsupportedKeepServer(t *testing.T) { + testManifestWithMultipleStreamsAndBlocks(t, 2, 2, "testblobstore", false) +} + +// Test datamanager with dry-run. Expect no block to be deleted. +func TestManifestWithMultipleStreamsAndBlocks_DryRun(t *testing.T) { + testManifestWithMultipleStreamsAndBlocks(t, 2, 2, "", true) +} + +func testManifestWithMultipleStreamsAndBlocks(t *testing.T, numStreams, numBlocks int, createExtraKeepServerWithType string, isDryRun bool) { + defer TearDownDataManagerTest(t) + SetupDataManagerTest(t) + + // create collection whose blocks will be backdated + collectionWithOldBlocks, oldBlocks := createMultiStreamBlockCollection(t, "old block", numStreams, numBlocks) + if collectionWithOldBlocks == "" { + t.Fatalf("Failed to create collection with %d blocks", numStreams*numBlocks) + } + if len(oldBlocks) != numStreams*numBlocks { + t.Fatalf("Not all blocks are created: expected %v, found %v", 1000, len(oldBlocks)) + } + + // create a stray block that will be backdated + strayOldBlock := putBlock(t, "this stray block is old") + + expected := []string{strayOldBlock} + expected = append(expected, oldBlocks...) + verifyBlocks(t, nil, expected, 2) + + // Backdate old blocks; but the collection still references these blocks + backdateBlocks(t, oldBlocks) + + // also backdate the stray old block + backdateBlocks(t, []string{strayOldBlock}) + + // If requested, create an extra keepserver with the given type + // This should be ignored during indexing and hence not change the datamanager outcome + var extraKeepServerUUID string + if createExtraKeepServerWithType != "" { + extraKeepServerUUID = addExtraKeepServer(t, createExtraKeepServerWithType) + defer deleteExtraKeepServer(extraKeepServerUUID) + } + + // run datamanager + dryRun = isDryRun + dataManagerSingleRun(t) + + if dryRun { + // verify that all blocks, including strayOldBlock, are still to be found + verifyBlocks(t, nil, expected, 2) + } else { + // verify that strayOldBlock is not to be found, but the collections blocks are still there + verifyBlocks(t, []string{strayOldBlock}, oldBlocks, 2) + } +} + +// Add one more keepstore with the given service type +func addExtraKeepServer(t *testing.T, serviceType string) string { + defer switchToken(arvadostest.AdminToken)() + + extraKeepService := make(arvadosclient.Dict) + err := arv.Create("keep_services", + arvadosclient.Dict{"keep_service": arvadosclient.Dict{ + "service_host": "localhost", + "service_port": "21321", + "service_ssl_flag": false, + "service_type": serviceType}}, + &extraKeepService) + if err != nil { + t.Fatal(err) + } + + return extraKeepService["uuid"].(string) +} + +func deleteExtraKeepServer(uuid string) { + defer switchToken(arvadostest.AdminToken)() + arv.Delete("keep_services", uuid, nil, nil) +}