prototype resumable send & recv

2017-08-18 23:52:33 +02:00
parent e5b713ce5b
commit d12ab6cd57
7 changed files with 234 additions and 3 deletions
@@ -87,6 +87,46 @@ func (h Handler) HandleIncrementalTransferRequest(r rpc.IncrementalTransferReque

 }

+func (h Handler) HandleResumeTransferRequest(r rpc.ResumeTransferRequest) (stream io.Reader, err error) {
+	// decode receive_resume_token: zfs send -nvP <token>
+	// A) use exit code to determine if could send (exit != 0 means could not send)
+	// B) check ACL if the filesystem in toname field (it is a snapshot so strip its snapshot name first)
+	//    is allowed for the given client
+	/*
+
+			# zfs send -nvt will print nvlist contents on ZoL and FreeBSD, should be good enough
+			# will print regardless of whether it can send or not -> good for us
+			# need clean way to extract that...
+				# expect 'resume token contents:\nnvlist version: 0\n'
+				# parse everything after that is tab-indented as key value pairs separated by = sign
+				# => return dict
+
+			zfs send -nv -t 1-c6491acbe-c8-789c636064000310a500c4ec50360710e72765a526973030b0419460caa7a515a79630c001489e0d493ea9b224b5182451885d7f497e7a69660a0343f79b1b9a8a2b3db65b20c97382e5f312735319188a4bf28b12d353f5931293b34b0b8af5ab8a520b72f4d3f2f31d8a53f35220660300c1091dbe
+		resume token contents:
+		nvlist version: 0
+		        object = 0x6
+		        offset = 0x0
+		        bytes = 0x7100
+		        toguid = 0xb748a92129d8ec8b
+		        toname = storage/backups/zrepl/foo@send
+		cannot resume send: 'storage/backups/zrepl/foo@send' used in the initial send no longer exists
+
+		zfs send -nvt 1-ebbbbea7e-f0-789c636064000310a501c49c50360710a715e5e7a69766a6304041f79b1b9a8a2b3db62b00d9ec48eaf293b252934b181858a0ea30e4d3d28a534b18e00024cf86249f5459925a0ca44fc861d75f920f71c59c5fdf6f7b3eea32b24092e704cbe725e6a632301497e41725a6a7ea27252667971614eb5715a516e4e8a7e5e73b14a7e6a51881cd06005a222749
+		resume token contents:
+		nvlist version: 0
+		        fromguid = 0xb748a92129d8ec8b #NOTE the fromguid field which is only in this one, so don't hardcode
+		        object = 0x4
+		        offset = 0x0
+		        bytes = 0x1ec8
+		        toguid = 0x328ae249dbf7fa9c
+		        toname = storage/backups/zrepl/foo@send2
+		send from storage/backups/zrepl/foo@send to storage/backups/zrepl/foo@send2 estimated size is 1.02M
+
+	*/
+	panic("not implemented")
+	return nil, nil
+}
+
 func (h Handler) HandlePullMeRequest(r rpc.PullMeRequest, clientIdentity string, client rpc.RPCRequester) (err error) {

 	// Check if we have a sink for this request
@@ -325,6 +325,87 @@ func doPull(pull PullContext) (err error) {
 			log("local filesystem does not exist")
 		case localState.Placeholder:
 			log("local filesystem is marked as placeholder")
+		case localState.ResumeToken != "":
+
+			log("local filesystem has receive_resume_token")
+			if false { // TODO Check if managed resume is disabled (explain information leakage in docs!)
+				log("managed resume tokens are disabled via zrepl config, assuming config change or external administrative action")
+				log("policy forbids aborting the partial recv automatically, skipping this filesystem")
+				log("for zrepl to resume replication for this dataset, administrators should finishing the recv manually or abort the recv via `zfs recv -A %s`", m.Local.ToString())
+				return true // TODO right choice to allow children? ... should be, since the fs exists...
+			}
+
+			log("decoding receive_resume_token")
+			// TODO, see handler comments
+			// TODO override logger with fromguid toguid
+
+			log("requesting resume of transfer")
+			r := rpc.ResumeTransferRequest{m.Remote, localState.ResumeToken}
+			stream, err := remote.ResumeTransferRequest(r)
+			if err != nil {
+
+				log("resume transfer request failed: %s", err)
+				rre, ok := err.(*rpc.ResumeTransferError)
+				if !ok {
+					log("skipping this filesystem, could be temporary issue")
+					return true // TODO right choice to allow children
+				}
+
+				// Determine if we should clear the resume token
+				clearAndUseSnaps := false
+				switch rre.Reason {
+				case rpc.ResumeTransferErrorReasonNotImplemented:
+					fallthrough
+				case rpc.ResumeTransferErrorReasonDisabled:
+					fallthrough
+				case rpc.ResumeTransferErrorReasonZFSErrorPermanent:
+					clearAndUseSnaps = true
+
+				case rpc.ResumeTransferErrorReasonZFSErrorMaybeTemporary:
+					fallthrough
+				default:
+					clearAndUseSnaps = false
+				}
+
+				if !clearAndUseSnaps {
+					log("skipping this filesystem, error identified as temporary")
+					return true // TODO right choice to allow children
+				}
+
+				log("clearing local receive_resume_token")
+				if err := zfs.ZFSRecvAbort(m.Local); err != nil {
+					log("error clearing receive_resume_token: %s", err)
+					return true // TODO right choice to allow children? the filesystem exists, so it should be ok
+				}
+				// TODO go back to top of function (put all this into separate function, then tail recursive call)
+				// TODO return this_function()
+			}
+
+			// TODO warning code duplication, see below. need to unify this
+			log("invoking zfs receive")
+			watcher := util.IOProgressWatcher{Reader: stream}
+			watcher.KickOff(1*time.Second, func(p util.IOProgress) {
+				log("progress on receive operation: %v bytes received", p.TotalRX)
+			})
+
+			recvArgs := []string{"-u"}
+			if localState.Placeholder {
+				log("receive with forced rollback to replace placeholder filesystem")
+				recvArgs = append(recvArgs, "-F")
+			}
+
+			if err = zfs.ZFSRecv(m.Local, &watcher, recvArgs...); err != nil {
+				log("error receiving stream: %s", err)
+				return false
+			}
+			log("finished receiving stream, %v bytes total", watcher.Progress().TotalRX)
+
+			// TODO further problem property handling code must be duplicated here...
+
+			// TODO tail recursive call to this function, we might not be dony syncing yet
+			// TODO return this_function()
+			return true
+
 		default:
 			log("local filesystem exists")
 			log("requesting local filesystem versions")