X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/668cf29c1a7f8270ed5ef0bcfbe693618a8155da..7cb15d038a126d4d16e6bfea670161baed0a1d7d:/services/datamanager/datamanager_test.go diff --git a/services/datamanager/datamanager_test.go b/services/datamanager/datamanager_test.go index 6ceb1e625e..28faf989ce 100644 --- a/services/datamanager/datamanager_test.go +++ b/services/datamanager/datamanager_test.go @@ -6,6 +6,8 @@ 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" @@ -16,11 +18,6 @@ import ( "time" ) -const ( - ActiveUserToken = "3kg6k6lzmp9kj5cpkcoxie963cmvjahbt2fod9zru30k1jqdmi" - AdminToken = "4axaw8zxe0qm22wa6urpp5nskcne8z88cvbupv653y1njyi05h" -) - var arv arvadosclient.ArvadosClient var keepClient *keepclient.KeepClient var keepServers []string @@ -34,10 +31,11 @@ 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) } + arv.ApiToken = arvadostest.DataManagerToken // keep client keepClient = &keepclient.KeepClient{ @@ -60,6 +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 { @@ -128,7 +128,18 @@ func getFirstLocatorFromCollection(t *testing.T, uuid string) string { return match[1] + "+" + match[2] } +func switchToken(t string) func() { + orig := arv.ApiToken + restore := func() { + arv.ApiToken = orig + } + arv.ApiToken = t + return restore +} + func getCollection(t *testing.T, uuid string) Dict { + defer switchToken(arvadostest.AdminToken)() + getback := make(Dict) err := arv.Get("collections", uuid, nil, &getback) if err != nil { @@ -142,6 +153,8 @@ func getCollection(t *testing.T, uuid string) Dict { } func updateCollection(t *testing.T, uuid string, paramName string, paramValue string) { + defer switchToken(arvadostest.AdminToken)() + err := arv.Update("collections", uuid, arvadosclient.Dict{ "collection": arvadosclient.Dict{ paramName: paramValue, @@ -156,6 +169,8 @@ func updateCollection(t *testing.T, uuid string, paramName string, paramValue st type Dict map[string]interface{} func deleteCollection(t *testing.T, uuid string) { + defer switchToken(arvadostest.AdminToken)() + getback := make(Dict) err := arv.Delete("collections", uuid, nil, &getback) if err != nil { @@ -179,7 +194,7 @@ func getBlockIndexesForServer(t *testing.T, i int) []string { path := keepServers[i] + "/index" client := http.Client{} req, err := http.NewRequest("GET", path, nil) - req.Header.Add("Authorization", "OAuth2 "+AdminToken) + req.Header.Add("Authorization", "OAuth2 "+arvadostest.DataManagerToken) req.Header.Add("Content-Type", "application/octet-stream") resp, err := client.Do(req) defer resp.Body.Close() @@ -243,13 +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 @@ -301,7 +313,7 @@ func backdateBlocks(t *testing.T, oldUnusedBlockLocators []string) { func getStatus(t *testing.T, path string) interface{} { client := http.Client{} req, err := http.NewRequest("GET", path, nil) - req.Header.Add("Authorization", "OAuth2 "+AdminToken) + req.Header.Add("Authorization", "OAuth2 "+arvadostest.DataManagerToken) req.Header.Add("Content-Type", "application/octet-stream") resp, err := client.Do(req) if err != nil { @@ -331,11 +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) @@ -393,6 +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...) @@ -401,6 +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) @@ -414,6 +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) @@ -426,6 +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) @@ -508,10 +526,133 @@ func TestRunDatamanagerAsNonAdminUser(t *testing.T) { defer TearDownDataManagerTest(t) SetupDataManagerTest(t) - arv.ApiToken = ActiveUserToken + arv.ApiToken = arvadostest.ActiveToken err := singlerun(arv) if err == nil { 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) { + testOldBlocksNotDeletedOnDataManagerError(t, "/badwritetofile", "", true, true) +} + +func TestPutAndGetBlocks_ErrorDuringGetCollectionsBadHeapProfileFilename(t *testing.T) { + testOldBlocksNotDeletedOnDataManagerError(t, "", "/badheapprofilefile", 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) { + defer TearDownDataManagerTest(t) + SetupDataManagerTest(t) + + // create collection whose blocks will be backdated + collectionWithOldBlocks, oldBlocks := createMultiStreamBlockCollection(t, "old block", 100, 10) + if collectionWithOldBlocks == "" { + t.Fatalf("Failed to create collection with 1000 blocks") + } + if len(oldBlocks) != 1000 { + 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}) + + // run datamanager + dataManagerSingleRun(t) + + // verify that strayOldBlock is not to be found, but the collections blocks are still there + verifyBlocks(t, []string{strayOldBlock}, oldBlocks, 2) +}