Merge pull request #51 from NETWAYS/feature/snapshot

Add subcommand for snapshots
NETWAYS · Apr 12, 2024 · ea9d866 · ea9d866
2 parents 4dc8764 + 39769b1
commit ea9d866
Show file tree

Hide file tree

Showing 5 changed files with 348 additions and 0 deletions.
diff --git a/README.md b/README.md
@@ -130,6 +130,46 @@ check_elasticsearch ingest --pipeline foobar
   \_[OK] Failed ingest operations for foobar: 5; | pipelines.foobar.failed=5c
 ```
 
+### Snapshot
+
+Checks status of Snapshots.
+
+```
+Checks the status of Elasticsearch snapshots
+The plugin maps snapshot status to the following status codes:
+
+SUCCESS, Exit code 0
+PARTIAL, Exit code 1
+FAILED, Exit code 2
+IN_PROGRESS, Exit code 3
+
+If there are multiple snapshots the plugin uses the worst status
+
+Usage:
+  check_elasticsearch snapshot [flags]
+
+Flags:
+  -a, --all                 Check all retrieved snapshots. If not set only the latest snapshot is checked
+  -N, --number int          Check latest N number snapshots. If not set only the latest snapshot is checked (default 1)
+  -r, --repository string   Comma-separated list of snapshot repository names used to limit the request (default "*")
+  -s, --snapshot string     Comma-separated list of snapshot names to retrieve. Wildcard (*) expressions are supported (default "*")
+  -h, --help                help for snapshot
+```
+
+Examples:
+
+```
+$ check_elasticsearch snapshot
+[OK] - All evaluated snapshots are in state SUCCESS
+
+$ check_elasticsearch snapshot --all -r myrepo
+[CRITICAL] - At least one evaluated snapshot is in state FAILED
+
+$ check_elasticsearch snapshot --number 5 -s mysnapshot
+[WARNING] - At least one evaluated snapshot is in state PARTIAL
+```
+
+
 ## License
 
 Copyright (c) 2022 [NETWAYS GmbH](mailto:[email protected])

diff --git a/cmd/snapshot.go b/cmd/snapshot.go
@@ -0,0 +1,126 @@
+package cmd
+
+import (
+	"fmt"
+	"strings"
+
+	"github.com/NETWAYS/go-check"
+	"github.com/NETWAYS/go-check/result"
+	"github.com/spf13/cobra"
+)
+
+var snapshotCmd = &cobra.Command{
+	Use:   "snapshot",
+	Short: "Checks the status of Elasticsearch snapshots",
+	Long: `Checks the status of Elasticsearch snapshots.
+The plugin maps snapshot status to the following exit codes:
+
+SUCCESS, Exit code 0
+PARTIAL, Exit code 1
+FAILED, Exit code 2
+IN_PROGRESS, Exit code 3
+
+If there are multiple snapshots the plugin uses the worst status.
+`,
+	Example: `
+$ check_elasticsearch snapshot
+[OK] - All evaluated snapshots are in state SUCCESS
+
+$ check_elasticsearch snapshot --all
+[CRITICAL] - At least one evaluated snapshot is in state FAILED
+
+$ check_elasticsearch snapshot --number 5
+[WARNING] - At least one evaluated snapshot is in state PARTIAL
+`,
+	Run: func(cmd *cobra.Command, args []string) {
+		repository, _ := cmd.Flags().GetString("repository")
+		snapshot, _ := cmd.Flags().GetString("snapshot")
+		numberOfSnapshots, _ := cmd.Flags().GetInt("number")
+		evalAllSnapshots, _ := cmd.Flags().GetBool("all")
+
+		var (
+			rc     int
+			output string
+		)
+
+		client := cliConfig.NewClient()
+
+		snapResponse, err := client.Snapshot(repository, snapshot)
+
+		if err != nil {
+			check.ExitError(err)
+		}
+
+		// If all snapshots are to be evaluated
+		if evalAllSnapshots {
+			numberOfSnapshots = len(snapResponse.Snapshots)
+		}
+
+		// If more snapshots are requested than available
+		if numberOfSnapshots > len(snapResponse.Snapshots) {
+			numberOfSnapshots = len(snapResponse.Snapshots)
+		}
+
+		// Evaluate snashots given their states
+		sStates := make([]int, 0, len(snapResponse.Snapshots))
+
+		// Check status for each snapshot
+		var summary strings.Builder
+
+		for _, snap := range snapResponse.Snapshots[:numberOfSnapshots] {
+
+			summary.WriteString("\n \\_")
+
+			switch snap.State {
+			default:
+				sStates = append(sStates, check.Unknown)
+				summary.WriteString(fmt.Sprintf("[UNKNOWN] Snapshot: %s, State %s, Repository: %s", snap.Snapshot, snap.State, snap.Repository))
+			case "SUCCESS":
+				sStates = append(sStates, check.OK)
+				summary.WriteString(fmt.Sprintf("[OK] Snapshot: %s, State %s, Repository: %s", snap.Snapshot, snap.State, snap.Repository))
+			case "PARTIAL":
+				sStates = append(sStates, check.Warning)
+				summary.WriteString(fmt.Sprintf("[WARNING] Snapshot: %s, State %s, Repository: %s", snap.Snapshot, snap.State, snap.Repository))
+			case "FAILED":
+				sStates = append(sStates, check.Critical)
+				summary.WriteString(fmt.Sprintf("[CRITICAL] Snapshot: %s, State %s, Repository: %s", snap.Snapshot, snap.State, snap.Repository))
+			case "IN PROGRESS":
+				sStates = append(sStates, check.Unknown)
+				summary.WriteString(fmt.Sprintf("[UNKNOWN] Snapshot: %s, State %s, Repository: %s", snap.Snapshot, snap.State, snap.Repository))
+			}
+		}
+
+		rc = result.WorstState(sStates...)
+
+		switch rc {
+		case check.OK:
+			output = "All evaluated snapshots are in state SUCCESS."
+		case check.Warning:
+			output = "At least one evaluated snapshot is in state PARTIAL."
+		case check.Critical:
+			output = "At least one evaluated snapshot is in state FAILED."
+		case check.Unknown:
+			output = "At least one evaluated snapshot is in state IN_PROGRESS."
+		default:
+			output = "Could not evaluate status of snapshots"
+		}
+
+		check.ExitRaw(rc, output, "repository:", repository, "snapshot:", snapshot, summary.String())
+	},
+}
+
+func init() {
+	rootCmd.AddCommand(snapshotCmd)
+
+	fs := snapshotCmd.Flags()
+
+	fs.StringP("snapshot", "s", "*",
+		"Comma-separated list of snapshot names to retrieve. Wildcard (*) expressions are supported")
+	fs.StringP("repository", "r", "*",
+		"Comma-separated list of snapshot repository names used to limit the request")
+
+	fs.IntP("number", "N", 1, "Check latest N number snapshots. If not set only the latest snapshot is checked")
+	fs.BoolP("all", "a", false, "Check all retrieved snapshots. If not set only the latest snapshot is checked")
+
+	snapshotCmd.MarkFlagsMutuallyExclusive("number", "all")
+}
diff --git a/cmd/snapshot_test.go b/cmd/snapshot_test.go
@@ -0,0 +1,120 @@
+package cmd
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"os/exec"
+	"strings"
+	"testing"
+)
+
+func TestSnapshot_ConnectionRefused(t *testing.T) {
+
+	cmd := exec.Command("go", "run", "../main.go", "snapshot", "--port", "9999")
+	out, _ := cmd.CombinedOutput()
+
+	actual := string(out)
+	expected := "[UNKNOWN] - could not fetch snapshots: Get \"http://localhost:9999/_snapshot/*/*?order=desc\": dial"
+
+	if !strings.Contains(actual, expected) {
+		t.Error("\nActual: ", actual, "\nExpected: ", expected)
+	}
+}
+
+func TestSnapshot_WithWrongFlags(t *testing.T) {
+
+	cmd := exec.Command("go", "run", "../main.go", "snapshot", "--all", "--number", "9999")
+	out, _ := cmd.CombinedOutput()
+
+	actual := string(out)
+	expected := "[UNKNOWN] - if any flags in the group"
+
+	if !strings.Contains(actual, expected) {
+		t.Error("\nActual: ", actual, "\nExpected: ", expected)
+	}
+}
+
+type SnapshotTest struct {
+	name     string
+	server   *httptest.Server
+	args     []string
+	expected string
+}
+
+func TestSnapshotCmd(t *testing.T) {
+	tests := []SnapshotTest{
+		{
+			name: "no-snapshot",
+			server: httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				w.Header().Set("X-Elastic-Product", "Elasticsearch")
+				w.WriteHeader(http.StatusOK)
+				w.Write([]byte(`Hey dude where my snapshot`))
+			})),
+			args:     []string{"run", "../main.go", "snapshot"},
+			expected: "[UNKNOWN] - could not decode snapshot response",
+		},
+		{
+			name: "snapshot-ok",
+			server: httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				w.Header().Set("X-Elastic-Product", "Elasticsearch")
+				w.WriteHeader(http.StatusOK)
+				w.Write([]byte(`{"snapshots":[{"snapshot":"snapshot_1","uuid":"dKb54xw67gvdRctLCxSket","repository":"my_repository","version_id":1.1,"version":1,"indices":[],"data_streams":[],"feature_states":[],"include_global_state":true,"state":"SUCCESS","start_time":"2020-07-06T21:55:18.129Z","start_time_in_millis":1593093628850,"end_time":"2020-07-06T21:55:18.129Z","end_time_in_millis":1593094752018,"duration_in_millis":0,"failures":[],"shards":{"total":0,"failed":0,"successful":0}},{"snapshot":"snapshot_2","uuid":"vdRctLCxSketdKb54xw67g","repository":"my_repository","version_id":2,"version":2,"indices":[],"data_streams":[],"feature_states":[],"include_global_state":true,"state":"SUCCESS","start_time":"2020-07-06T21:55:18.130Z","start_time_in_millis":1593093628851,"end_time":"2020-07-06T21:55:18.130Z","end_time_in_millis":1593094752019,"duration_in_millis":1,"failures":[],"shards":{"total":0,"failed":0,"successful":0}}],"next":"c25hcHNob3RfMixteV9yZXBvc2l0b3J5LHNuYXBzaG90XzI=","total":3,"remaining":1}
+`))
+			})),
+			args:     []string{"run", "../main.go", "snapshot"},
+			expected: "[OK] - All evaluated snapshots are in state SUCCESS",
+		},
+		{
+			name: "snapshot-inprogress",
+			server: httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				w.Header().Set("X-Elastic-Product", "Elasticsearch")
+				w.WriteHeader(http.StatusOK)
+				w.Write([]byte(`{"snapshots":[{"snapshot":"snapshot_1","uuid":"dKb54xw67gvdRctLCxSket","repository":"my_repository","version_id":1,"version":1,"indices":[],"data_streams":[],"feature_states":[],"include_global_state":true,"state":"IN_PROGRESS","start_time":"2020-07-06T21:55:18.129Z","start_time_in_millis":1593093628850,"end_time":"2020-07-06T21:55:18.129Z","end_time_in_millis":1593094752018,"duration_in_millis":0,"failures":[],"shards":{"total":0,"failed":0,"successful":0}},{"snapshot":"snapshot_2","uuid":"vdRctLCxSketdKb54xw67g","repository":"my_repository","version_id":2,"version":2,"indices":[],"data_streams":[],"feature_states":[],"include_global_state":true,"state":"SUCCESS","start_time":"2020-07-06T21:55:18.130Z","start_time_in_millis":1593093628851,"end_time":"2020-07-06T21:55:18.130Z","end_time_in_millis":1593094752019,"duration_in_millis":1,"failures":[],"shards":{"total":0,"failed":0,"successful":0}}],"next":"c25hcHNob3RfMixteV9yZXBvc2l0b3J5LHNuYXBzaG90XzI=","total":3,"remaining":1}
+`))
+			})),
+			args:     []string{"run", "../main.go", "snapshot"},
+			expected: "[UNKNOWN] - At least one evaluated snapshot is in state IN_PROGRESS",
+		},
+		{
+			name: "snapshot-failed-with-all",
+			server: httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				w.Header().Set("X-Elastic-Product", "Elasticsearch")
+				w.WriteHeader(http.StatusOK)
+				w.Write([]byte(`{"snapshots":[{"snapshot":"snapshot_1","uuid":"dKb54xw67gvdRctLCxSket","repository":"my_repository","version_id":1,"version":1,"indices":[],"data_streams":[],"feature_states":[],"include_global_state":true,"state":"IN_PROGRESS","start_time":"2020-07-06T21:55:18.129Z","start_time_in_millis":1593093628850,"end_time":"2020-07-06T21:55:18.129Z","end_time_in_millis":1593094752018,"duration_in_millis":0,"failures":[],"shards":{"total":0,"failed":0,"successful":0}},{"snapshot":"snapshot_2","uuid":"vdRctLCxSketdKb54xw67g","repository":"my_repository","version_id":2,"version":2,"indices":[],"data_streams":[],"feature_states":[],"include_global_state":true,"state":"FAILED","start_time":"2020-07-06T21:55:18.130Z","start_time_in_millis":1593093628851,"end_time":"2020-07-06T21:55:18.130Z","end_time_in_millis":1593094752019,"duration_in_millis":1,"failures":[],"shards":{"total":0,"failed":0,"successful":0}}],"next":"c25hcHNob3RfMixteV9yZXBvc2l0b3J5LHNuYXBzaG90XzI=","total":3,"remaining":1}
+`))
+			})),
+			args:     []string{"run", "../main.go", "snapshot", "--all"},
+			expected: "[CRITICAL] - At least one evaluated snapshot is in state FAILED",
+		},
+		{
+			name: "snapshot-partial-with-number",
+			server: httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				w.Header().Set("X-Elastic-Product", "Elasticsearch")
+				w.WriteHeader(http.StatusOK)
+				w.Write([]byte(`{"snapshots":[{"snapshot":"snapshot_1","uuid":"dKb54xw67gvdRctLCxSket","repository":"my_repository","version_id":1,"version":1,"indices":[],"data_streams":[],"feature_states":[],"include_global_state":true,"state":"SUCCESS","start_time":"2020-07-06T21:55:18.129Z","start_time_in_millis":1593093628850,"end_time":"2020-07-06T21:55:18.129Z","end_time_in_millis":1593094752018,"duration_in_millis":0,"failures":[],"shards":{"total":0,"failed":0,"successful":0}},{"snapshot":"snapshot_2","uuid":"vdRctLCxSketdKb54xw67g","repository":"my_repository","version_id":2,"version":2,"indices":[],"data_streams":[],"feature_states":[],"include_global_state":true,"state":"PARTIAL","start_time":"2020-07-06T21:55:18.130Z","start_time_in_millis":1593093628851,"end_time":"2020-07-06T21:55:18.130Z","end_time_in_millis":1593094752019,"duration_in_millis":1,"failures":[],"shards":{"total":0,"failed":0,"successful":0}}],"next":"c25hcHNob3RfMixteV9yZXBvc2l0b3J5LHNuYXBzaG90XzI=","total":3,"remaining":1}
+`))
+			})),
+			args:     []string{"run", "../main.go", "snapshot", "--number", "4"},
+			expected: "[WARNING] - At least one evaluated snapshot is in state PARTIAL",
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			defer test.server.Close()
+
+			// We need the random Port extracted
+			u, _ := url.Parse(test.server.URL)
+			cmd := exec.Command("go", append(test.args, "--port", u.Port())...)
+			out, _ := cmd.CombinedOutput()
+
+			actual := string(out)
+
+			if !strings.Contains(actual, test.expected) {
+				t.Error("\nActual: ", actual, "\nExpected: ", test.expected)
+			}
+
+		})
+	}
+}
diff --git a/internal/client/client.go b/internal/client/client.go
@@ -179,3 +179,39 @@ func (c *Client) NodeStats() (r *es.ClusterStats, err error) {
 
 	return
 }
+
+func (c *Client) Snapshot(repository string, snapshot string) (*es.SnapshotResponse, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+
+	r := &es.SnapshotResponse{}
+
+	u, _ := url.JoinPath(c.URL, "/_snapshot/", repository, snapshot)
+
+	// Retrieve snapshots in descending order to get latest
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, u+"?order=desc", nil)
+
+	if err != nil {
+		return r, fmt.Errorf("error creating request: %w", err)
+	}
+
+	resp, err := c.Client.Do(req)
+
+	if err != nil {
+		return r, fmt.Errorf("could not fetch snapshots: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		return r, fmt.Errorf("request failed for snapshots: %s", resp.Status)
+	}
+
+	defer resp.Body.Close()
+
+	err = json.NewDecoder(resp.Body).Decode(r)
+
+	if err != nil {
+		return r, fmt.Errorf("could not decode snapshot response: %w", err)
+	}
+
+	return r, nil
+}
diff --git a/internal/elasticsearch/api.go b/internal/elasticsearch/api.go
@@ -79,3 +79,29 @@ type ClusterStats struct {
 	Nodes       map[string]NodeInfo `json:"nodes"`
 	ClusterName string              `json:"cluster_name"`
 }
+
+type Snapshot struct {
+	Snapshot           string   `json:"snapshot"`
+	UUID               string   `json:"uuid"`
+	Repository         string   `json:"repository"`
+	Indices            []string `json:"indices"`
+	DataStreams        []string `json:"data_streams"`
+	FeatureStates      []string `json:"feature_states"`
+	IncludeGlobalState bool     `json:"include_global_state"`
+	State              string   `json:"state"`
+	StartTimeInMillis  int      `json:"start_time_in_millis"`
+	EndTimeInMillis    int      `json:"end_time_in_millis"`
+	DurationInMillis   int      `json:"duration_in_millis"`
+	Failures           []string `json:"failures"`
+	Shards             struct {
+		Total      int `json:"total"`
+		Failed     int `json:"failed"`
+		Successful int `json:"successful"`
+	} `json:"shards"`
+}
+
+type SnapshotResponse struct {
+	Snapshots []Snapshot `json:"snapshots"`
+	Total     int        `json:"total"`
+	Remaining int        `json:"remaining"`
+}