X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/4df1175e30c21850af394fcd60c9bb7ca3d981a5..5824ee2e5198dd46a7813fe2adbd380a114f9ac4:/services/datamanager/collection/collection.go diff --git a/services/datamanager/collection/collection.go b/services/datamanager/collection/collection.go index 424db83562..e0929b7757 100644 --- a/services/datamanager/collection/collection.go +++ b/services/datamanager/collection/collection.go @@ -1,4 +1,4 @@ -/* Deals with parsing Collection responses from API Server. */ +// Deals with parsing Collection responses from API Server. package collection @@ -13,18 +13,21 @@ import ( "git.curoverse.com/arvados.git/services/datamanager/loggerutil" "log" "os" - "runtime" "runtime/pprof" "time" ) var ( - heap_profile_filename string + heapProfileFilename string // globals for debugging totalManifestSize uint64 maxManifestSize uint64 ) +const ( + DefaultReplicationLevel = 2 +) + type Collection struct { Uuid string OwnerUuid string @@ -34,9 +37,13 @@ type Collection struct { } type ReadCollections struct { - ReadAllCollections bool - UuidToCollection map[string]Collection - OwnerToCollectionSize map[string]int + ReadAllCollections bool + UuidToCollection map[string]Collection + OwnerToCollectionSize map[string]int + BlockToReplication map[blockdigest.BlockDigest]int + CollectionUuidToIndex map[string]int + CollectionIndexToUuid []string + BlockToCollectionIndices map[blockdigest.BlockDigest][]int } type GetCollectionsParams struct { @@ -59,21 +66,12 @@ type SdkCollectionList struct { } func init() { - flag.StringVar(&heap_profile_filename, + flag.StringVar(&heapProfileFilename, "heap-profile", "", "File to write the heap profiles to. Leave blank to skip profiling.") } -// // Methods to implement util.SdkListResponse Interface -// func (s SdkCollectionList) NumItemsAvailable() (numAvailable int, err error) { -// return s.ItemsAvailable, nil -// } - -// func (s SdkCollectionList) NumItemsContained() (numContained int, err error) { -// return len(s.Items), nil -// } - // Write the heap profile to a file for later review. // Since a file is expected to only contain a single heap profile this // function overwrites the previously written profile, so it is safe @@ -81,9 +79,9 @@ func init() { // Otherwise we would see cumulative numbers as explained here: // https://groups.google.com/d/msg/golang-nuts/ZyHciRglQYc/2nh4Ndu2fZcJ func WriteHeapProfile() { - if heap_profile_filename != "" { + if heapProfileFilename != "" { - heap_profile, err := os.Create(heap_profile_filename) + heap_profile, err := os.Create(heapProfileFilename) if err != nil { log.Fatal(err) } @@ -99,17 +97,7 @@ func WriteHeapProfile() { func GetCollectionsAndSummarize(params GetCollectionsParams) (results ReadCollections) { results = GetCollections(params) - ComputeSizeOfOwnedCollections(&results) - - if params.Logger != nil { - params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) { - collectionInfo := p["collection_info"].(map[string]interface{}) - // Since maps are shallow copied, we run a risk of concurrent - // updates here. By copying results.OwnerToCollectionSize into - // the log, we're assuming that it won't be updated. - collectionInfo["owner_to_collection_size"] = results.OwnerToCollectionSize - }) - } + results.Summarize(params.Logger) log.Printf("Uuid to Size used: %v", results.OwnerToCollectionSize) log.Printf("Read and processed %d collections", @@ -134,7 +122,6 @@ func GetCollections(params GetCollectionsParams) (results ReadCollections) { fieldsWanted := []string{"manifest_text", "owner_uuid", "uuid", - // TODO(misha): Start using the redundancy field. "redundancy", "modified_at"} @@ -161,10 +148,9 @@ func GetCollections(params GetCollectionsParams) (results ReadCollections) { if params.Logger != nil { params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) { - collectionInfo := make(map[string]interface{}) + collectionInfo := logger.GetOrCreateMap(p, "collection_info") collectionInfo["num_collections_at_start"] = initialNumberOfCollectionsAvailable collectionInfo["batch_size"] = params.BatchSize - p["collection_info"] = collectionInfo }) } @@ -206,7 +192,7 @@ func GetCollections(params GetCollectionsParams) (results ReadCollections) { if params.Logger != nil { params.Logger.Update(func(p map[string]interface{}, e map[string]interface{}) { - collectionInfo := p["collection_info"].(map[string]interface{}) + collectionInfo := logger.GetOrCreateMap(p, "collection_info") collectionInfo["collections_read"] = totalCollections collectionInfo["latest_modified_date_seen"] = sdkParams["filters"].([][]string)[0][2] collectionInfo["total_manifest_size"] = totalManifestSize @@ -215,9 +201,6 @@ func GetCollections(params GetCollectionsParams) (results ReadCollections) { } } - // Just in case this lowers the numbers reported in the heap profile. - runtime.GC() - // Write the heap profile for examining memory usage WriteHeapProfile() @@ -244,15 +227,20 @@ func ProcessCollections(arvLogger *logger.Logger, loggerutil.FatalWithMessage(arvLogger, fmt.Sprintf( "Arvados SDK collection returned with unexpected zero "+ - "modifcation date. This probably means that either we failed to "+ + "modification date. This probably means that either we failed to "+ "parse the modification date or the API server has changed how "+ - "it returns modification dates: %v", + "it returns modification dates: %+v", collection)) } if sdkCollection.ModifiedAt.After(latestModificationDate) { latestModificationDate = sdkCollection.ModifiedAt } + + if collection.ReplicationLevel == 0 { + collection.ReplicationLevel = DefaultReplicationLevel + } + manifest := manifest.Manifest{sdkCollection.ManifestText} manifestSize := uint64(len(sdkCollection.ManifestText)) @@ -291,11 +279,44 @@ func ProcessCollections(arvLogger *logger.Logger, return } -func ComputeSizeOfOwnedCollections(readCollections *ReadCollections) { +func (readCollections *ReadCollections) Summarize(arvLogger *logger.Logger) { readCollections.OwnerToCollectionSize = make(map[string]int) + readCollections.BlockToReplication = make(map[blockdigest.BlockDigest]int) + numCollections := len(readCollections.UuidToCollection) + readCollections.CollectionUuidToIndex = make(map[string]int, numCollections) + readCollections.CollectionIndexToUuid = make([]string, 0, numCollections) + readCollections.BlockToCollectionIndices = make(map[blockdigest.BlockDigest][]int) + for _, coll := range readCollections.UuidToCollection { + collectionIndex := len(readCollections.CollectionIndexToUuid) + readCollections.CollectionIndexToUuid = + append(readCollections.CollectionIndexToUuid, coll.Uuid) + readCollections.CollectionUuidToIndex[coll.Uuid] = collectionIndex + readCollections.OwnerToCollectionSize[coll.OwnerUuid] = readCollections.OwnerToCollectionSize[coll.OwnerUuid] + coll.TotalSize + + for block, _ := range coll.BlockDigestToSize { + readCollections.BlockToCollectionIndices[block] = + append(readCollections.BlockToCollectionIndices[block], collectionIndex) + storedReplication := readCollections.BlockToReplication[block] + if coll.ReplicationLevel > storedReplication { + readCollections.BlockToReplication[block] = coll.ReplicationLevel + } + } + } + + if arvLogger != nil { + arvLogger.Update(func(p map[string]interface{}, e map[string]interface{}) { + collectionInfo := logger.GetOrCreateMap(p, "collection_info") + // Since maps are shallow copied, we run a risk of concurrent + // updates here. By copying results.OwnerToCollectionSize into + // the log, we're assuming that it won't be updated. + collectionInfo["owner_to_collection_size"] = + readCollections.OwnerToCollectionSize + collectionInfo["distinct_blocks_named"] = + len(readCollections.BlockToReplication) + }) } return