// Blocks outgoing network traffic on 'node'. Then verifies that 'podNameToDisappear',
// that belongs to replication controller 'rcName', really disappeared.
// Finally, it checks that the replication controller recreates the
// pods on another node and that now the number of replicas is equal 'replicas'.
// At the end (even in case of errors), the network traffic is brought back to normal.
// This function executes commands on a node so it will work only for some
// environments.
func performTemporaryNetworkFailure(c *client.Client, ns, rcName string, replicas int32, podNameToDisappear string, node *api.Node) {
host := getNodeExternalIP(node)
master := getMaster(c)
By(fmt.Sprintf("block network traffic from node %s to the master", node.Name))
defer func() {
// This code will execute even if setting the iptables rule failed.
// It is on purpose because we may have an error even if the new rule
// had been inserted. (yes, we could look at the error code and ssh error
// separately, but I prefer to stay on the safe side).
By(fmt.Sprintf("Unblock network traffic from node %s to the master", node.Name))
framework.UnblockNetwork(host, master)
}()
framework.Logf("Waiting %v to ensure node %s is ready before beginning test...", resizeNodeReadyTimeout, node.Name)
if !framework.WaitForNodeToBe(c, node.Name, api.NodeReady, true, resizeNodeReadyTimeout) {
framework.Failf("Node %s did not become ready within %v", node.Name, resizeNodeReadyTimeout)
}
framework.BlockNetwork(host, master)
framework.Logf("Waiting %v for node %s to be not ready after simulated network failure", resizeNodeNotReadyTimeout, node.Name)
if !framework.WaitForNodeToBe(c, node.Name, api.NodeReady, false, resizeNodeNotReadyTimeout) {
framework.Failf("Node %s did not become not-ready within %v", node.Name, resizeNodeNotReadyTimeout)
}
framework.Logf("Waiting for pod %s to be removed", podNameToDisappear)
err := framework.WaitForRCPodToDisappear(c, ns, rcName, podNameToDisappear)
Expect(err).NotTo(HaveOccurred())
By("verifying whether the pod from the unreachable node is recreated")
err = framework.VerifyPods(c, ns, rcName, true, replicas)
Expect(err).NotTo(HaveOccurred())
// network traffic is unblocked in a deferred function
}
// Blocks outgoing network traffic on 'node'. Then runs testFunc and returns its status.
// At the end (even in case of errors), the network traffic is brought back to normal.
// This function executes commands on a node so it will work only for some
// environments.
func testUnderTemporaryNetworkFailure(c clientset.Interface, ns string, node *api.Node, testFunc func()) {
host := framework.GetNodeExternalIP(node)
master := framework.GetMasterAddress(c)
By(fmt.Sprintf("block network traffic from node %s to the master", node.Name))
defer func() {
// This code will execute even if setting the iptables rule failed.
// It is on purpose because we may have an error even if the new rule
// had been inserted. (yes, we could look at the error code and ssh error
// separately, but I prefer to stay on the safe side).
By(fmt.Sprintf("Unblock network traffic from node %s to the master", node.Name))
framework.UnblockNetwork(host, master)
}()
framework.Logf("Waiting %v to ensure node %s is ready before beginning test...", resizeNodeReadyTimeout, node.Name)
if !framework.WaitForNodeToBe(c, node.Name, api.NodeReady, true, resizeNodeReadyTimeout) {
framework.Failf("Node %s did not become ready within %v", node.Name, resizeNodeReadyTimeout)
}
framework.BlockNetwork(host, master)
framework.Logf("Waiting %v for node %s to be not ready after simulated network failure", resizeNodeNotReadyTimeout, node.Name)
if !framework.WaitForNodeToBe(c, node.Name, api.NodeReady, false, resizeNodeNotReadyTimeout) {
framework.Failf("Node %s did not become not-ready within %v", node.Name, resizeNodeNotReadyTimeout)
}
testFunc()
// network traffic is unblocked in a deferred function
}
// recoverDiskSpace recovers disk space, filled by creating a large file, on a given node.
func recoverDiskSpace(c *client.Client, node *api.Node) {
By(fmt.Sprintf("Recovering disk space on node %s", node.Name))
cmd := "rm -f test.img"
framework.ExpectNoError(framework.IssueSSHCommand(cmd, framework.TestContext.Provider, node))
ood := framework.WaitForNodeToBe(c, node.Name, api.NodeOutOfDisk, false, nodeOODTimeOut)
Expect(ood).To(BeTrue(), "Node %s's out of disk condition status did not change to false within %v", node.Name, nodeOODTimeOut)
}
// fillDiskSpace fills the available disk space on a given node by creating a large file. The disk
// space on the node is filled in such a way that the available space after filling the disk is just
// below the lowDiskSpaceThreshold mark.
func fillDiskSpace(c *client.Client, node *api.Node) {
avail, err := availSize(c, node)
framework.ExpectNoError(err, "Node %s: couldn't obtain available disk size %v", node.Name, err)
fillSize := (avail - lowDiskSpaceThreshold + (100 * mb))
framework.Logf("Node %s: disk space available %d bytes", node.Name, avail)
By(fmt.Sprintf("Node %s: creating a file of size %d bytes to fill the available disk space", node.Name, fillSize))
cmd := fmt.Sprintf("fallocate -l %d test.img", fillSize)
framework.ExpectNoError(framework.IssueSSHCommand(cmd, framework.TestContext.Provider, node))
ood := framework.WaitForNodeToBe(c, node.Name, api.NodeOutOfDisk, true, nodeOODTimeOut)
Expect(ood).To(BeTrue(), "Node %s did not run out of disk within %v", node.Name, nodeOODTimeOut)
avail, err = availSize(c, node)
framework.Logf("Node %s: disk space available %d bytes", node.Name, avail)
Expect(avail < lowDiskSpaceThreshold).To(BeTrue())
}