Merge pull request #316 from ystia/bugfix/GH-315_fix_panic_ssh_pool

Fixed panic in case of ssh connection error

CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ## UNRELEASED
 
+### BUG FIXES
+
+* SSH Session pool: Panic if connection failed, this impacts Slurm infrastructure ([GH-315](https://github.com/ystia/yorc/issues/315))
+
 ## 3.1.1 (February 06, 2019)
 
 ### BUG FIXES
@@ -10,13 +14,13 @@
 * Can't deploy applications using a secured yorc/consul ([GH-274](https://github.com/ystia/yorc/issues/274))
 * Unable to purge an application that appears in the list ([GH-238](https://github.com/ystia/yorc/issues/238))
 * K8S jobs namespace should not be removed if its provided ([GH-245](https://github.com/ystia/yorc/issues/245))
-* Deployment with a topology parsing error remains in initial status ([GH-283](https://github.com/ystia/yorc/issues/283)
-* Interface name is not retrieved from custom command Rest request ([GH-287](https://github.com/ystia/yorc/issues/287)
-* Instances are adding into topology before creating task ([GH-289](https://github.com/ystia/yorc/issues/289)
-* All ssh connections to Slurm are killed if ssh server has reached the max number of allowed sessions ([GH-291](https://github.com/ystia/yorc/issues/291)
+* Deployment with a topology parsing error remains in initial status ([GH-283](https://github.com/ystia/yorc/issues/283))
+* Interface name is not retrieved from custom command Rest request ([GH-287](https://github.com/ystia/yorc/issues/287))
+* Instances are adding into topology before creating task ([GH-289](https://github.com/ystia/yorc/issues/289))
+* All ssh connections to Slurm are killed if ssh server has reached the max number of allowed sessions ([GH-291](https://github.com/ystia/yorc/issues/291))
 * Missing events for uninstall workflow in purge task ([GH-302](https://github.com/ystia/yorc/issues/302)
-* It can take a considerable delay for a deployment to change status to UNDEPLOYMENT_IN_PROGRESS ([GH-306](https://github.com/ystia/yorc/issues/306)
-* Slurm job monitoring is not designed for concurrency ([GH-308](https://github.com/ystia/yorc/issues/308)
+* It can take a considerable delay for a deployment to change status to UNDEPLOYMENT_IN_PROGRESS ([GH-306](https://github.com/ystia/yorc/issues/306))
+* Slurm job monitoring is not designed for concurrency ([GH-308](https://github.com/ystia/yorc/issues/308))
 
 ## 3.1.0 (December 20, 2018)
 

helper/sshutil/session_pool.go

@@ -146,8 +146,17 @@ func (p *pool) getConn(k, addr string, config *ssh.ClientConfig) *conn {
 	p.tab[k] = c
 	p.mu.Unlock()
 	c.netC, c.c, c.err = p.dial("tcp", addr, config)
-	c.name = fmt.Sprintf("%s-%s-%x", addr, config.User, c.c.SessionID())
-	metrics.IncrCounter(metricsutil.CleanupMetricKey([]string{"ssh-connections-pool", "creations", c.name}), 1)
+	// Add context to the error if any
+	c.err = errors.Wrapf(c.err, "failed to open connection on %s", addr)
+	if c.err == nil {
+		// Connection succeeded
+		c.name = fmt.Sprintf("%s-%s-%x", addr, config.User, c.c.SessionID())
+		metrics.IncrCounter(metricsutil.CleanupMetricKey([]string{"ssh-connections-pool", "creations", c.name}), 1)
+	} else {
+		// Connection failed
+		c.name = fmt.Sprintf("%s-%s", addr, config.User)
+		metrics.IncrCounter(metricsutil.CleanupMetricKey([]string{"ssh-connections-pool", "create-failed", c.name}), 1)
+	}
 	close(c.ok)
 	return c
 }