//
// SPDX-License-Identifier: AGPL-3.0
-package main
+package keepstore
import (
"container/list"
"strconv"
"strings"
"sync"
+ "sync/atomic"
"time"
+ "git.arvados.org/arvados.git/lib/cmd"
"git.arvados.org/arvados.git/sdk/go/arvados"
"git.arvados.org/arvados.git/sdk/go/ctxlog"
"git.arvados.org/arvados.git/sdk/go/health"
}
func (rtr *router) handleGET(resp http.ResponseWriter, req *http.Request) {
- ctx, cancel := contextForResponse(context.TODO(), resp)
- defer cancel()
-
locator := req.URL.Path[1:]
if strings.Contains(locator, "+R") && !strings.Contains(locator, "+A") {
- rtr.remoteProxy.Get(ctx, resp, req, rtr.cluster, rtr.volmgr)
+ rtr.remoteProxy.Get(req.Context(), resp, req, rtr.cluster, rtr.volmgr)
return
}
// isn't here, we can return 404 now instead of waiting for a
// buffer.
- buf, err := getBufferWithContext(ctx, bufs, BlockSize)
+ buf, err := getBufferWithContext(req.Context(), bufs, BlockSize)
if err != nil {
http.Error(resp, err.Error(), http.StatusServiceUnavailable)
return
}
defer bufs.Put(buf)
- size, err := GetBlock(ctx, rtr.volmgr, mux.Vars(req)["hash"], buf, resp)
+ size, err := GetBlock(req.Context(), rtr.volmgr, mux.Vars(req)["hash"], buf, resp)
if err != nil {
code := http.StatusInternalServerError
if err, ok := err.(*KeepError); ok {
resp.Write(buf[:size])
}
-// Return a new context that gets cancelled by resp's CloseNotifier.
-func contextForResponse(parent context.Context, resp http.ResponseWriter) (context.Context, context.CancelFunc) {
- ctx, cancel := context.WithCancel(parent)
- if cn, ok := resp.(http.CloseNotifier); ok {
- go func(c <-chan bool) {
- select {
- case <-c:
- cancel()
- case <-ctx.Done():
- }
- }(cn.CloseNotify())
- }
- return ctx, cancel
-}
-
// Get a buffer from the pool -- but give up and return a non-nil
// error if ctx ends before we get a buffer.
func getBufferWithContext(ctx context.Context, bufs *bufferPool, bufSize int) ([]byte, error) {
}
func (rtr *router) handlePUT(resp http.ResponseWriter, req *http.Request) {
- ctx, cancel := contextForResponse(context.TODO(), resp)
- defer cancel()
-
hash := mux.Vars(req)["hash"]
// Detect as many error conditions as possible before reading
return
}
- buf, err := getBufferWithContext(ctx, bufs, int(req.ContentLength))
+ var wantStorageClasses []string
+ if hdr := req.Header.Get("X-Keep-Storage-Classes"); hdr != "" {
+ wantStorageClasses = strings.Split(hdr, ",")
+ for i, sc := range wantStorageClasses {
+ wantStorageClasses[i] = strings.TrimSpace(sc)
+ }
+ } else {
+ // none specified -- use configured default
+ for class, cfg := range rtr.cluster.StorageClasses {
+ if cfg.Default {
+ wantStorageClasses = append(wantStorageClasses, class)
+ }
+ }
+ }
+
+ buf, err := getBufferWithContext(req.Context(), bufs, int(req.ContentLength))
if err != nil {
http.Error(resp, err.Error(), http.StatusServiceUnavailable)
return
return
}
- replication, err := PutBlock(ctx, rtr.volmgr, buf, hash)
+ result, err := PutBlock(req.Context(), rtr.volmgr, buf, hash, wantStorageClasses)
bufs.Put(buf)
if err != nil {
expiry := time.Now().Add(rtr.cluster.Collections.BlobSigningTTL.Duration())
returnHash = SignLocator(rtr.cluster, returnHash, apiToken, expiry)
}
- resp.Header().Set("X-Keep-Replicas-Stored", strconv.Itoa(replication))
+ resp.Header().Set("X-Keep-Replicas-Stored", result.TotalReplication())
+ resp.Header().Set("X-Keep-Storage-Classes-Confirmed", result.ClassReplication())
resp.Write([]byte(returnHash + "\n"))
}
// populate the given NodeStatus struct with current values.
func (rtr *router) readNodeStatus(st *NodeStatus) {
- st.Version = version
+ st.Version = strings.SplitN(cmd.Version.String(), " ", 2)[0]
vols := rtr.volmgr.AllReadable()
if cap(st.Volumes) < len(vols) {
st.Volumes = make([]*volumeStatusEnt, len(vols))
if filehash != hash {
// TODO: Try harder to tell a sysadmin about
// this.
- log.Error("checksum mismatch for block %s (actual %s) on %s", hash, filehash, vol)
+ log.Errorf("checksum mismatch for block %s (actual %s), size %d on %s", hash, filehash, size, vol)
errorToCaller = DiskHashError
continue
}
return 0, errorToCaller
}
-// PutBlock Stores the BLOCK (identified by the content id HASH) in Keep.
-//
-// PutBlock(ctx, block, hash)
-// Stores the BLOCK (identified by the content id HASH) in Keep.
-//
-// The MD5 checksum of the block must be identical to the content id HASH.
-// If not, an error is returned.
+type putProgress struct {
+ classNeeded map[string]bool
+ classTodo map[string]bool
+ mountUsed map[*VolumeMount]bool
+ totalReplication int
+ classDone map[string]int
+}
+
+// Number of distinct replicas stored. "2" can mean the block was
+// stored on 2 different volumes with replication 1, or on 1 volume
+// with replication 2.
+func (pr putProgress) TotalReplication() string {
+ return strconv.Itoa(pr.totalReplication)
+}
+
+// Number of replicas satisfying each storage class, formatted like
+// "default=2; special=1".
+func (pr putProgress) ClassReplication() string {
+ s := ""
+ for k, v := range pr.classDone {
+ if len(s) > 0 {
+ s += ", "
+ }
+ s += k + "=" + strconv.Itoa(v)
+ }
+ return s
+}
+
+func (pr *putProgress) Add(mnt *VolumeMount) {
+ if pr.mountUsed[mnt] {
+ logrus.Warnf("BUG? superfluous extra write to mount %s", mnt.UUID)
+ return
+ }
+ pr.mountUsed[mnt] = true
+ pr.totalReplication += mnt.Replication
+ for class := range mnt.StorageClasses {
+ pr.classDone[class] += mnt.Replication
+ delete(pr.classTodo, class)
+ }
+}
+
+func (pr *putProgress) Sub(mnt *VolumeMount) {
+ if !pr.mountUsed[mnt] {
+ logrus.Warnf("BUG? Sub called with no prior matching Add: %s", mnt.UUID)
+ return
+ }
+ pr.mountUsed[mnt] = false
+ pr.totalReplication -= mnt.Replication
+ for class := range mnt.StorageClasses {
+ pr.classDone[class] -= mnt.Replication
+ if pr.classNeeded[class] {
+ pr.classTodo[class] = true
+ }
+ }
+}
+
+func (pr *putProgress) Done() bool {
+ return len(pr.classTodo) == 0 && pr.totalReplication > 0
+}
+
+func (pr *putProgress) Want(mnt *VolumeMount) bool {
+ if pr.Done() || pr.mountUsed[mnt] {
+ return false
+ }
+ if len(pr.classTodo) == 0 {
+ // none specified == "any"
+ return true
+ }
+ for class := range mnt.StorageClasses {
+ if pr.classTodo[class] {
+ return true
+ }
+ }
+ return false
+}
+
+func (pr *putProgress) Copy() *putProgress {
+ cp := putProgress{
+ classNeeded: pr.classNeeded,
+ classTodo: make(map[string]bool, len(pr.classTodo)),
+ classDone: make(map[string]int, len(pr.classDone)),
+ mountUsed: make(map[*VolumeMount]bool, len(pr.mountUsed)),
+ totalReplication: pr.totalReplication,
+ }
+ for k, v := range pr.classTodo {
+ cp.classTodo[k] = v
+ }
+ for k, v := range pr.classDone {
+ cp.classDone[k] = v
+ }
+ for k, v := range pr.mountUsed {
+ cp.mountUsed[k] = v
+ }
+ return &cp
+}
+
+func newPutProgress(classes []string) putProgress {
+ pr := putProgress{
+ classNeeded: make(map[string]bool, len(classes)),
+ classTodo: make(map[string]bool, len(classes)),
+ classDone: map[string]int{},
+ mountUsed: map[*VolumeMount]bool{},
+ }
+ for _, c := range classes {
+ if c != "" {
+ pr.classNeeded[c] = true
+ pr.classTodo[c] = true
+ }
+ }
+ return pr
+}
+
+// PutBlock stores the given block on one or more volumes.
//
-// PutBlock stores the BLOCK on the first Keep volume with free space.
-// A failure code is returned to the user only if all volumes fail.
+// The MD5 checksum of the block must match the given hash.
//
-// On success, PutBlock returns nil.
-// On failure, it returns a KeepError with one of the following codes:
+// The block is written to each writable volume (ordered by priority
+// and then UUID, see volume.go) until at least one replica has been
+// stored in each of the requested storage classes.
//
-// 500 Collision
-// A different block with the same hash already exists on this
-// Keep server.
-// 422 MD5Fail
-// The MD5 hash of the BLOCK does not match the argument HASH.
-// 503 Full
-// There was not enough space left in any Keep volume to store
-// the object.
-// 500 Fail
-// The object could not be stored for some other reason (e.g.
-// all writes failed). The text of the error message should
-// provide as much detail as possible.
+// The returned error, if any, is a KeepError with one of the
+// following codes:
//
-func PutBlock(ctx context.Context, volmgr *RRVolumeManager, block []byte, hash string) (int, error) {
+// 500 Collision
+// A different block with the same hash already exists on this
+// Keep server.
+// 422 MD5Fail
+// The MD5 hash of the BLOCK does not match the argument HASH.
+// 503 Full
+// There was not enough space left in any Keep volume to store
+// the object.
+// 500 Fail
+// The object could not be stored for some other reason (e.g.
+// all writes failed). The text of the error message should
+// provide as much detail as possible.
+func PutBlock(ctx context.Context, volmgr *RRVolumeManager, block []byte, hash string, wantStorageClasses []string) (putProgress, error) {
log := ctxlog.FromContext(ctx)
// Check that BLOCK's checksum matches HASH.
blockhash := fmt.Sprintf("%x", md5.Sum(block))
if blockhash != hash {
log.Printf("%s: MD5 checksum %s did not match request", hash, blockhash)
- return 0, RequestHashError
+ return putProgress{}, RequestHashError
}
+ result := newPutProgress(wantStorageClasses)
+
// If we already have this data, it's intact on disk, and we
// can update its timestamp, return success. If we have
// different data with the same hash, return failure.
- if n, err := CompareAndTouch(ctx, volmgr, hash, block); err == nil || err == CollisionError {
- return n, err
- } else if ctx.Err() != nil {
- return 0, ErrClientDisconnect
- }
-
- // Choose a Keep volume to write to.
- // If this volume fails, try all of the volumes in order.
- if mnt := volmgr.NextWritable(); mnt != nil {
- if err := mnt.Put(ctx, hash, block); err != nil {
- log.WithError(err).Errorf("%s: Put(%s) failed", mnt.Volume, hash)
- } else {
- return mnt.Replication, nil // success!
- }
+ if err := CompareAndTouch(ctx, volmgr, hash, block, &result); err != nil || result.Done() {
+ return result, err
}
if ctx.Err() != nil {
- return 0, ErrClientDisconnect
+ return result, ErrClientDisconnect
}
- writables := volmgr.AllWritable()
+ writables := volmgr.NextWritable()
if len(writables) == 0 {
log.Error("no writable volumes")
- return 0, FullError
+ return result, FullError
}
- allFull := true
- for _, vol := range writables {
- err := vol.Put(ctx, hash, block)
- if ctx.Err() != nil {
- return 0, ErrClientDisconnect
+ var wg sync.WaitGroup
+ var mtx sync.Mutex
+ cond := sync.Cond{L: &mtx}
+ // pending predicts what result will be if all pending writes
+ // succeed.
+ pending := result.Copy()
+ var allFull atomic.Value
+ allFull.Store(true)
+
+ // We hold the lock for the duration of the "each volume" loop
+ // below, except when it is released during cond.Wait().
+ mtx.Lock()
+
+ for _, mnt := range writables {
+ // Wait until our decision to use this mount does not
+ // depend on the outcome of pending writes.
+ for result.Want(mnt) && !pending.Want(mnt) {
+ cond.Wait()
}
- switch err {
- case nil:
- return vol.Replication, nil // success!
- case FullError:
+ if !result.Want(mnt) {
continue
- default:
- // The volume is not full but the
- // write did not succeed. Report the
- // error and continue trying.
- allFull = false
- log.WithError(err).Errorf("%s: Put(%s) failed", vol, hash)
}
+ mnt := mnt
+ pending.Add(mnt)
+ wg.Add(1)
+ go func() {
+ log.Debugf("PutBlock: start write to %s", mnt.UUID)
+ defer wg.Done()
+ err := mnt.Put(ctx, hash, block)
+
+ mtx.Lock()
+ if err != nil {
+ log.Debugf("PutBlock: write to %s failed", mnt.UUID)
+ pending.Sub(mnt)
+ } else {
+ log.Debugf("PutBlock: write to %s succeeded", mnt.UUID)
+ result.Add(mnt)
+ }
+ cond.Broadcast()
+ mtx.Unlock()
+
+ if err != nil && err != FullError && ctx.Err() == nil {
+ // The volume is not full but the
+ // write did not succeed. Report the
+ // error and continue trying.
+ allFull.Store(false)
+ log.WithError(err).Errorf("%s: Put(%s) failed", mnt.Volume, hash)
+ }
+ }()
+ }
+ mtx.Unlock()
+ wg.Wait()
+ if ctx.Err() != nil {
+ return result, ErrClientDisconnect
+ }
+ if result.Done() {
+ return result, nil
}
- if allFull {
- log.Error("all volumes are full")
- return 0, FullError
+ if result.totalReplication > 0 {
+ // Some, but not all, of the storage classes were
+ // satisfied. This qualifies as success.
+ return result, nil
+ } else if allFull.Load().(bool) {
+ log.Error("all volumes with qualifying storage classes are full")
+ return putProgress{}, FullError
+ } else {
+ // Already logged the non-full errors.
+ return putProgress{}, GenericError
}
- // Already logged the non-full errors.
- return 0, GenericError
}
-// CompareAndTouch returns the current replication level if one of the
-// volumes already has the given content and it successfully updates
-// the relevant block's modification time in order to protect it from
-// premature garbage collection. Otherwise, it returns a non-nil
-// error.
-func CompareAndTouch(ctx context.Context, volmgr *RRVolumeManager, hash string, buf []byte) (int, error) {
+// CompareAndTouch looks for volumes where the given content already
+// exists and its modification time can be updated (i.e., it is
+// protected from garbage collection), and updates result accordingly.
+// It returns when the result is Done() or all volumes have been
+// checked.
+func CompareAndTouch(ctx context.Context, volmgr *RRVolumeManager, hash string, buf []byte, result *putProgress) error {
log := ctxlog.FromContext(ctx)
- var bestErr error = NotFoundError
for _, mnt := range volmgr.AllWritable() {
+ if !result.Want(mnt) {
+ continue
+ }
err := mnt.Compare(ctx, hash, buf)
if ctx.Err() != nil {
- return 0, ctx.Err()
+ return nil
} else if err == CollisionError {
// Stop if we have a block with same hash but
// different content. (It will be impossible
// to tell which one is wanted if we have
// both, so there's no point writing it even
// on a different volume.)
- log.Error("collision in Compare(%s) on volume %s", hash, mnt.Volume)
- return 0, err
+ log.Errorf("collision in Compare(%s) on volume %s", hash, mnt.Volume)
+ return CollisionError
} else if os.IsNotExist(err) {
// Block does not exist. This is the only
// "normal" error: we don't log anything.
}
if err := mnt.Touch(hash); err != nil {
log.WithError(err).Errorf("error in Touch(%s) on volume %s", hash, mnt.Volume)
- bestErr = err
continue
}
// Compare and Touch both worked --> done.
- return mnt.Replication, nil
+ result.Add(mnt)
+ if result.Done() {
+ return nil
+ }
}
- return 0, bestErr
+ return nil
}
var validLocatorRe = regexp.MustCompile(`^[0-9a-f]{32}$`)
projects / arvados.git / blobdiff