|
| 1 | +package node |
| 2 | + |
| 3 | +import ( |
| 4 | + "fmt" |
| 5 | + "regexp" |
| 6 | + "strconv" |
| 7 | + "strings" |
| 8 | + "sync" |
| 9 | + |
| 10 | + g "github.com/onsi/ginkgo/v2" |
| 11 | + o "github.com/onsi/gomega" |
| 12 | + ote "github.com/openshift-eng/openshift-tests-extension/pkg/ginkgo" |
| 13 | + exutil "github.com/openshift/origin/test/extended/util" |
| 14 | +) |
| 15 | + |
| 16 | +// goroutineHeaderRe matches Go goroutine dump headers, e.g.: |
| 17 | +// |
| 18 | +// goroutine 64418 [IO wait, 189 minutes]: |
| 19 | +// goroutine 1 [running]: |
| 20 | +var goroutineHeaderRe = regexp.MustCompile(`goroutine \d+ \[`) |
| 21 | + |
| 22 | +// ioWaitDurationRe extracts the wait duration in minutes from an IO wait goroutine header, e.g.: |
| 23 | +// |
| 24 | +// goroutine 64418 [IO wait, 189 minutes]: |
| 25 | +var ioWaitDurationRe = regexp.MustCompile(`goroutine \d+ \[IO wait, (\d+) minutes\]:`) |
| 26 | + |
| 27 | +// findStuckImagePulls checks a goroutine dump for goroutines that indicate |
| 28 | +// a stuck image pull: IO wait > 30 minutes with net.(*conn).Read called by |
| 29 | +// image/v5/docker.(*bodyReader).Read. |
| 30 | +func findStuckImagePulls(dump string) []string { |
| 31 | + // Split the dump into individual goroutine blocks. |
| 32 | + // Each block starts with "goroutine <id> [". |
| 33 | + blocks := goroutineHeaderRe.Split(dump, -1) |
| 34 | + headers := goroutineHeaderRe.FindAllString(dump, -1) |
| 35 | + |
| 36 | + var stuck []string |
| 37 | + for i, header := range headers { |
| 38 | + block := header + blocks[i+1] |
| 39 | + |
| 40 | + // Check: state is "IO wait" with duration > 30 minutes |
| 41 | + match := ioWaitDurationRe.FindStringSubmatch(block) |
| 42 | + if match == nil { |
| 43 | + continue |
| 44 | + } |
| 45 | + minutes, err := strconv.Atoi(match[1]) |
| 46 | + if err != nil || minutes <= 30 { |
| 47 | + continue |
| 48 | + } |
| 49 | + |
| 50 | + // Check: stack contains net.(*conn).Read |
| 51 | + connReadIdx := strings.Index(block, "net.(*conn).Read") |
| 52 | + if connReadIdx < 0 { |
| 53 | + continue |
| 54 | + } |
| 55 | + |
| 56 | + // Check: image/v5/docker.(*bodyReader).Read appears as an |
| 57 | + // ascendant (caller) of net.(*conn).Read. In a goroutine dump |
| 58 | + // callers appear below callees, so bodyReader must come after |
| 59 | + // conn.Read in the text. |
| 60 | + bodyReaderIdx := strings.Index(block, "image/v5/docker.(*bodyReader).Read") |
| 61 | + if bodyReaderIdx < 0 || bodyReaderIdx <= connReadIdx { |
| 62 | + continue |
| 63 | + } |
| 64 | + |
| 65 | + stuck = append(stuck, strings.TrimSpace(block)) |
| 66 | + } |
| 67 | + return stuck |
| 68 | +} |
| 69 | + |
| 70 | +var _ = g.Describe("[sig-node][Late]", func() { |
| 71 | + defer g.GinkgoRecover() |
| 72 | + oc := exutil.NewCLIWithoutNamespace("crio-goroutine-dump") |
| 73 | + |
| 74 | + g.It("CRI-O should report goroutine stacks on all nodes", |
| 75 | + ote.Informing(), func(ctx g.SpecContext) { |
| 76 | + |
| 77 | + nodes, err := exutil.GetAllClusterNodes(oc) |
| 78 | + o.Expect(err).NotTo(o.HaveOccurred()) |
| 79 | + o.Expect(nodes).NotTo(o.BeEmpty(), "expected at least one node") |
| 80 | + |
| 81 | + // Send SIGUSR1 to CRI-O and read the newly created dump file. |
| 82 | + // CRI-O writes goroutine stacks to /tmp/crio-goroutine-stacks-<timestamp>.log. |
| 83 | + // |
| 84 | + // We access CRI-O's /tmp through /proc/<pid>/root/tmp rather than |
| 85 | + // the host /tmp so that the dump is visible even when CRI-O's |
| 86 | + // systemd unit uses PrivateTmp=yes (as observed on MicroShift). |
| 87 | + // |
| 88 | + // The script snapshots the latest dump file before signalling and |
| 89 | + // polls until a newer, non-empty (-s) file appears to avoid |
| 90 | + // reading a file that CRI-O has created but not yet finished writing. |
| 91 | + shellCmd := `CRIO_PID=$(pgrep -x crio 2>/dev/null) |
| 92 | +if [ -z "$CRIO_PID" ]; then echo "CRIO_NOT_FOUND"; exit 0; fi |
| 93 | +CRIO_TMP=/proc/$CRIO_PID/root/tmp |
| 94 | +BEFORE=$(ls -t "$CRIO_TMP"/crio-goroutine-stacks-*.log 2>/dev/null | head -1) |
| 95 | +kill -USR1 $CRIO_PID |
| 96 | +for i in $(seq 1 30); do |
| 97 | + LATEST=$(ls -t "$CRIO_TMP"/crio-goroutine-stacks-*.log 2>/dev/null | head -1) |
| 98 | + if [ -n "$LATEST" ] && [ "$LATEST" != "$BEFORE" ] && [ -s "$LATEST" ]; then |
| 99 | + cat "$LATEST"; exit 0 |
| 100 | + fi |
| 101 | + sleep 1 |
| 102 | +done |
| 103 | +echo "DUMP_TIMEOUT"; exit 1` |
| 104 | + |
| 105 | + // nodeResult holds the output from a single node's goroutine dump. |
| 106 | + type nodeResult struct { |
| 107 | + name string |
| 108 | + output string |
| 109 | + err error |
| 110 | + stuckPulls []string |
| 111 | + } |
| 112 | + |
| 113 | + results := make([]nodeResult, len(nodes)) |
| 114 | + var wg sync.WaitGroup |
| 115 | + for i, node := range nodes { |
| 116 | + wg.Add(1) |
| 117 | + go func(i int, nodeName string) { |
| 118 | + defer g.GinkgoRecover() |
| 119 | + defer wg.Done() |
| 120 | + |
| 121 | + g.By(fmt.Sprintf("Sending SIGUSR1 to CRI-O on node %s", nodeName)) |
| 122 | + |
| 123 | + output, err := exutil.DebugNodeRetryWithOptionsAndChroot( |
| 124 | + oc, nodeName, "default", |
| 125 | + "sh", "-c", shellCmd, |
| 126 | + ) |
| 127 | + results[i] = nodeResult{name: nodeName, output: output, err: err} |
| 128 | + }(i, node.Name) |
| 129 | + } |
| 130 | + wg.Wait() |
| 131 | + |
| 132 | + var dumps []string |
| 133 | + var stuckPulls []string |
| 134 | + var failedNodes []string |
| 135 | + for _, r := range results { |
| 136 | + // Check output-based diagnostics before the generic error |
| 137 | + // because DebugNodeRetryWithOptionsAndChroot may return |
| 138 | + // both output and an error (e.g. non-zero exit from the |
| 139 | + // DUMP_TIMEOUT branch). |
| 140 | + if strings.Contains(r.output, "CRIO_NOT_FOUND") { |
| 141 | + failedNodes = append(failedNodes, |
| 142 | + fmt.Sprintf("%s: CRI-O process not found", r.name)) |
| 143 | + continue |
| 144 | + } |
| 145 | + |
| 146 | + if strings.Contains(r.output, "DUMP_TIMEOUT") { |
| 147 | + failedNodes = append(failedNodes, |
| 148 | + fmt.Sprintf("%s: timed out waiting for new goroutine dump file", r.name)) |
| 149 | + continue |
| 150 | + } |
| 151 | + |
| 152 | + if r.err != nil { |
| 153 | + failedNodes = append(failedNodes, |
| 154 | + fmt.Sprintf("%s: debug pod failed: %v", r.name, r.err)) |
| 155 | + continue |
| 156 | + } |
| 157 | + |
| 158 | + o.Expect(goroutineHeaderRe.MatchString(r.output)).To(o.BeTrue(), |
| 159 | + "expected goroutine stacks in CRI-O dump from node %s, output length=%d, got:\n%s", r.name, len(r.output), r.output) |
| 160 | + |
| 161 | + dumps = append(dumps, fmt.Sprintf("=== node/%s ===\n%s", r.name, r.output)) |
| 162 | + |
| 163 | + for _, goroutine := range findStuckImagePulls(r.output) { |
| 164 | + stuckPulls = append(stuckPulls, fmt.Sprintf("node/%s:\n%s", r.name, goroutine)) |
| 165 | + } |
| 166 | + } |
| 167 | + |
| 168 | + // Fail if any node did not produce a dump |
| 169 | + o.Expect(failedNodes).To(o.BeEmpty(), |
| 170 | + "failed to collect CRI-O goroutine dump from nodes:\n%s", |
| 171 | + strings.Join(failedNodes, "\n")) |
| 172 | + |
| 173 | + // Fail hard if any goroutine is stuck in an image pull |
| 174 | + o.Expect(stuckPulls).To(o.BeEmpty(), |
| 175 | + "found CRI-O goroutines stuck in image pull (IO wait > 30 min):\n%s", |
| 176 | + strings.Join(stuckPulls, "\n\n")) |
| 177 | + |
| 178 | + // Always fail so goroutine dumps are visible in test results. |
| 179 | + // The test is marked Informing so this won't block merges. |
| 180 | + if len(dumps) > 0 { |
| 181 | + g.Fail(fmt.Sprintf("CRI-O goroutine dumps collected:\n%s", strings.Join(dumps, "\n\n"))) |
| 182 | + } |
| 183 | + }) |
| 184 | +}) |
0 commit comments