X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/75e196b201be758ffb857c3f2d33847e8cca0d18..50128b53da4003912635b03fb27b5be2c5beaca1:/services/keepstore/volume.go diff --git a/services/keepstore/volume.go b/services/keepstore/volume.go index cdaec92326..b72258d51a 100644 --- a/services/keepstore/volume.go +++ b/services/keepstore/volume.go @@ -1,6 +1,7 @@ package main import ( + "context" "io" "sync/atomic" "time" @@ -10,17 +11,23 @@ import ( // for example, a single mounted disk, a RAID array, an Amazon S3 volume, // etc. type Volume interface { - // Get a block. IFF the returned error is nil, the caller must - // put the returned slice back into the buffer pool when it's - // finished with it. (Otherwise, the buffer pool will be - // depleted and eventually -- when all available buffers are - // used and not returned -- operations will reach deadlock.) + // Volume type as specified in config file. Examples: "S3", + // "Directory". + Type() string + + // Do whatever private setup tasks and configuration checks + // are needed. Return non-nil if the volume is unusable (e.g., + // invalid config). + Start() error + + // Get a block: copy the block data into buf, and return the + // number of bytes copied. // // loc is guaranteed to consist of 32 or more lowercase hex // digits. // - // Get should not verify the integrity of the returned data: - // it should just return whatever was found in its backing + // Get should not verify the integrity of the data: it should + // just return whatever was found in its backing // store. (Integrity checking is the caller's responsibility.) // // If an error is encountered that prevents it from @@ -36,23 +43,25 @@ type Volume interface { // access log if the block is not found on any other volumes // either). // - // If the data in the backing store is bigger than BLOCKSIZE, - // Get is permitted to return an error without reading any of - // the data. - Get(loc string) ([]byte, error) + // If the data in the backing store is bigger than len(buf), + // then Get is permitted to return an error without reading + // any of the data. + // + // len(buf) will not exceed BlockSize. + Get(ctx context.Context, loc string, buf []byte) (int, error) // Compare the given data with the stored data (i.e., what Get // would return). If equal, return nil. If not, return // CollisionError or DiskHashError (depending on whether the // data on disk matches the expected hash), or whatever error // was encountered opening/reading the stored data. - Compare(loc string, data []byte) error + Compare(ctx context.Context, loc string, data []byte) error // Put writes a block to an underlying storage device. // // loc is as described in Get. // - // len(block) is guaranteed to be between 0 and BLOCKSIZE. + // len(block) is guaranteed to be between 0 and BlockSize. // // If a block is already stored under the same name (loc) with // different content, Put must either overwrite the existing @@ -76,7 +85,7 @@ type Volume interface { // // Put should not verify that loc==hash(block): this is the // caller's responsibility. - Put(loc string, block []byte) error + Put(ctx context.Context, loc string, block []byte) error // Touch sets the timestamp for the given locator to the // current time. @@ -144,20 +153,21 @@ type Volume interface { // particular order. IndexTo(prefix string, writer io.Writer) error - // Delete deletes the block data from the underlying storage - // device. + // Trash moves the block data from the underlying storage + // device to trash area. The block then stays in trash for + // -trash-lifetime interval before it is actually deleted. // // loc is as described in Get. // // If the timestamp for the given locator is newer than - // blob_signature_ttl, Delete must not delete the data. + // BlobSignatureTTL, Trash must not trash the data. // - // If a Delete operation overlaps with any Touch or Put + // If a Trash operation overlaps with any Touch or Put // operations on the same locator, the implementation must // ensure one of the following outcomes: // // - Touch and Put return a non-nil error, or - // - Delete does not delete the block, or + // - Trash does not trash the block, or // - Both of the above. // // If it is possible for the storage device to be accessed by @@ -171,9 +181,12 @@ type Volume interface { // reliably or fail outright. // // Corollary: A successful Touch or Put guarantees a block - // will not be deleted for at least blob_signature_ttl + // will not be trashed for at least BlobSignatureTTL // seconds. - Delete(loc string) error + Trash(loc string) error + + // Untrash moves block from trash back into store + Untrash(loc string) error // Status returns a *VolumeStatus representing the current // in-use and available storage capacity and an @@ -195,6 +208,22 @@ type Volume interface { // will fail because it is full, but Mtime or Delete can // succeed -- then Writable should return false. Writable() bool + + // Replication returns the storage redundancy of the + // underlying device. It will be passed on to clients in + // responses to PUT requests. + Replication() int + + // EmptyTrash looks for trashed blocks that exceeded TrashLifetime + // and deletes them from the volume. + EmptyTrash() +} + +// A VolumeWithExamples provides example configs to display in the +// -help message. +type VolumeWithExamples interface { + Volume + Examples() []Volume } // A VolumeManager tells callers which volumes can read, which volumes @@ -214,6 +243,10 @@ type VolumeManager interface { // with more free space, etc. NextWritable() Volume + // VolumeStats returns the ioStats used for tracking stats for + // the given Volume. + VolumeStats(Volume) *ioStats + // Close shuts down the volume manager cleanly. Close() } @@ -225,12 +258,16 @@ type RRVolumeManager struct { readables []Volume writables []Volume counter uint32 + iostats map[Volume]*ioStats } // MakeRRVolumeManager initializes RRVolumeManager func MakeRRVolumeManager(volumes []Volume) *RRVolumeManager { - vm := &RRVolumeManager{} + vm := &RRVolumeManager{ + iostats: make(map[Volume]*ioStats), + } for _, v := range volumes { + vm.iostats[v] = &ioStats{} vm.readables = append(vm.readables, v) if v.Writable() { vm.writables = append(vm.writables, v) @@ -258,18 +295,35 @@ func (vm *RRVolumeManager) NextWritable() Volume { return vm.writables[i%uint32(len(vm.writables))] } +// VolumeStats returns an ioStats for the given volume. +func (vm *RRVolumeManager) VolumeStats(v Volume) *ioStats { + return vm.iostats[v] +} + // Close the RRVolumeManager func (vm *RRVolumeManager) Close() { } -// VolumeStatus provides status information of the volume consisting of: -// * mount_point -// * device_num (an integer identifying the underlying storage system) -// * bytes_free -// * bytes_used +// VolumeStatus describes the current condition of a volume type VolumeStatus struct { - MountPoint string `json:"mount_point"` - DeviceNum uint64 `json:"device_num"` - BytesFree uint64 `json:"bytes_free"` - BytesUsed uint64 `json:"bytes_used"` + MountPoint string + DeviceNum uint64 + BytesFree uint64 + BytesUsed uint64 +} + +// ioStats tracks I/O statistics for a volume or server +type ioStats struct { + Errors uint64 + Ops uint64 + CompareOps uint64 + GetOps uint64 + PutOps uint64 + TouchOps uint64 + InBytes uint64 + OutBytes uint64 +} + +type InternalStatser interface { + InternalStats() interface{} }