ssh+stdinserver: connect: dial_timeout

This  is a follow-up to ccd062e

Gopkg.lock

@@ -83,7 +83,7 @@
   branch = "master"
   name = "github.com/problame/go-netssh"
   packages = ["."]
-  revision = "ffa145d2506e222977205e7666a9722d6b9959ac"
+  revision = "984ce91f575c75c4d5d6c7e74660d2c837571496"
 
 [[projects]]
   branch = "master"

cmd/config_connect.go

@@ -9,6 +9,7 @@ import (
 	"github.com/mitchellh/mapstructure"
 	"github.com/pkg/errors"
 	"github.com/problame/go-netssh"
+	"time"
 )
 
 type SSHStdinserverConnecter struct {
@@ -19,6 +20,8 @@ type SSHStdinserverConnecter struct {
 	TransportOpenCommand []string `mapstructure:"transport_open_command"`
 	SSHCommand           string   `mapstructure:"ssh_command"`
 	Options              []string
+	DialTimeout          string `mapstructure:"dial_timeout"`
+	dialTimeout          time.Duration
 }
 
 func parseSSHStdinserverConnecter(i map[string]interface{}) (c *SSHStdinserverConnecter, err error) {
@@ -29,6 +32,15 @@ func parseSSHStdinserverConnecter(i map[string]interface{}) (c *SSHStdinserverCo
 		return nil, err
 	}
 
+	if c.DialTimeout != "" {
+		c.dialTimeout, err = time.ParseDuration(c.DialTimeout)
+		if err != nil {
+			return nil, errors.Wrap(err, "cannot parse dial_timeout")
+		}
+	} else {
+		c.dialTimeout = 10 * time.Second
+	}
+
 	// TODO assert fields are filled
 	return
 
@@ -38,9 +50,15 @@ func (c *SSHStdinserverConnecter) Connect() (rwc io.ReadWriteCloser, err error)
 
 	var endpoint netssh.Endpoint
 	if err = copier.Copy(&endpoint, c); err != nil {
-		return
+		return nil, errors.WithStack(err)
 	}
-	if rwc, err = netssh.Dial(context.TODO(), endpoint); err != nil {
+	var dialCtx context.Context
+	dialCtx, dialCancel := context.WithTimeout(context.TODO(), c.dialTimeout) // context.TODO tied to error handling below
+	defer dialCancel()
+	if rwc, err = netssh.Dial(dialCtx, endpoint); err != nil {
+		if err == context.DeadlineExceeded {
+			err = errors.Errorf("dial_timeout of %s exceeded", c.dialTimeout)
+		}
 		err = errors.WithStack(err)
 		return
 	}

docs/changelog.rst

@@ -19,7 +19,13 @@ Developers should consult the git commit log or GitHub issue tracker.
   * Make sure to understand the meaning bookmarks have for :ref:`maximum replication downtime <replication-downtime>`.
   * Example: :sampleconf:`pullbackup/productionhost.yml`
 
-* |break| :commit:`ccd062e`: both sides of a replication setup must be updated and restarted. Otherwise the connecting side will hang and not time out.
+* |break| :commit:`ccd062e`: ``ssh+stdinserver`` transport: changed protocol requires daemon restart on both sides
+
+  * The delicate procedure of talking to the serving-side zrepl daemon via the stdinserver proxy command now has better error handling.
+  * This includes handshakes between client+proxy and client + remote daemo, which is not implemented in previous versions of zrepl.
+  * The connecting side will therefore time out, with the message ``dial_timeout of 10s exceeded``.
+  * Both sides of a replication setup must be updated and restarted. Otherwise the connecting side will hang and not time out.
+
 * |break_config| :commit:`2bfcfa5`: first outlet in ``global.logging`` is now used for logging meta-errors, for example problems encountered when writing to other outlets.
 * |feature| :issue:`10`: ``zrepl control status`` subcommand
 

docs/configuration/transports.rst

@@ -85,8 +85,9 @@ Connect Mode
         user: root
         port: 22
         identity_file: /etc/zrepl/ssh/identity
-        options: # optional
+        options: # optional, default [], `-o` arguments passed to ssh
         - "Compression=on"
+        dial_timeout: # optional, default 10s, max time.Duration until initial handshake is completed
 
 The connecting zrepl daemon