19377: Add diagnostics -v flag: arvados/nginx/config versions, etc.

Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@curii.com>

diff --git a/lib/diagnostics/cmd.go b/lib/diagnostics/cmd.go
index 3a2ebe0c280bf68f6e7e397e65489c70196f91ae..5799ceb9a98d880ce0b5c7193a2707c43e12c042 100644 (file)
--- a/lib/diagnostics/cmd.go
+++ b/lib/diagnostics/cmd.go
@@ -38,6 +38,7 @@ func (Command) RunCommand(prog string, args []string, stdin io.Reader, stdout, s
        f.StringVar(&diag.dockerImage, "docker-image", "", "image to use when running a test container (default: use embedded hello-world image)")
        f.BoolVar(&diag.checkInternal, "internal-client", false, "check that this host is considered an \"internal\" client")
        f.BoolVar(&diag.checkExternal, "external-client", false, "check that this host is considered an \"external\" client")
+       f.BoolVar(&diag.verbose, "v", false, "verbose: include more information in report")
        f.IntVar(&diag.priority, "priority", 500, "priority for test container (1..1000, or 0 to skip)")
        f.DurationVar(&diag.timeout, "timeout", 10*time.Second, "timeout for http requests")
        if ok, code := cmd.ParseFlags(f, prog, args, "", stderr); !ok {
@@ -73,6 +74,7 @@ type diagnoser struct {
        dockerImage   string
        checkInternal bool
        checkExternal bool
+       verbose       bool
        timeout       time.Duration
        logger        *logrus.Logger
        errors        []string
@@ -87,6 +89,12 @@ func (diag *diagnoser) infof(f string, args ...interface{}) {
        diag.logger.Infof("  ... "+f, args...)
 }
 
+func (diag *diagnoser) verbosef(f string, args ...interface{}) {
+       if diag.verbose {
+               diag.logger.Infof("  ... "+f, args...)
+       }
+}
+
 func (diag *diagnoser) warnf(f string, args ...interface{}) {
        diag.logger.Warnf("  ... "+f, args...)
 }
@@ -128,6 +136,13 @@ func (diag *diagnoser) runtests() {
                return
        }
 
+       hostname, err := os.Hostname()
+       if err != nil {
+               diag.warnf("error getting hostname: %s")
+       } else {
+               diag.verbosef("hostname = %s", hostname)
+       }
+
        diag.dotest(5, "running health check (same as `arvados-server check`)", func() error {
                ldr := config.NewLoader(&bytes.Buffer{}, ctxlog.New(&bytes.Buffer{}, "text", "info"))
                ldr.SetupFlags(flag.NewFlagSet("diagnostics", flag.ContinueOnError))
@@ -148,7 +163,29 @@ func (diag *diagnoser) runtests() {
                for _, e := range resp.Errors {
                        diag.errorf("health check: %s", e)
                }
-               diag.infof("health check: reported clock skew %v", resp.ClockSkew)
+               diag.verbosef("reported clock skew = %v", resp.ClockSkew)
+               reported := map[string]bool{}
+               for _, result := range resp.Checks {
+                       version := strings.SplitN(result.Metrics.Version, " (go", 2)[0]
+                       if version != "" && !reported[version] {
+                               diag.verbosef("arvados version = %s", version)
+                               reported[version] = true
+                       }
+               }
+               reported = map[string]bool{}
+               for _, result := range resp.Checks {
+                       if result.Server != "" && !reported[result.Server] {
+                               diag.verbosef("http frontend version = %s", result.Server)
+                               reported[result.Server] = true
+                       }
+               }
+               reported = map[string]bool{}
+               for _, result := range resp.Checks {
+                       if sha := result.ConfigSourceSHA256; sha != "" && !reported[sha] {
+                               diag.verbosef("config file sha256 = %s", sha)
+                               reported[sha] = true
+                       }
+               }
                return nil
        })
 
@@ -161,7 +198,7 @@ func (diag *diagnoser) runtests() {
                if err != nil {
                        return err
                }
-               diag.debugf("BlobSignatureTTL = %d", dd.BlobSignatureTTL)
+               diag.verbosef("BlobSignatureTTL = %d", dd.BlobSignatureTTL)
                return nil
        })
 
@@ -175,7 +212,7 @@ func (diag *diagnoser) runtests() {
                if err != nil {
                        return err
                }
-               diag.debugf("Collections.BlobSigning = %v", cluster.Collections.BlobSigning)
+               diag.verbosef("Collections.BlobSigning = %v", cluster.Collections.BlobSigning)
                cfgOK = true
                return nil
        })
@@ -188,7 +225,7 @@ func (diag *diagnoser) runtests() {
                if err != nil {
                        return err
                }
-               diag.debugf("user uuid = %s", user.UUID)
+               diag.verbosef("user uuid = %s", user.UUID)
                return nil
        })
 
@@ -277,9 +314,9 @@ func (diag *diagnoser) runtests() {
                isInternal := found["proxy"] == 0 && len(keeplist.Items) > 0
                isExternal := found["proxy"] > 0 && found["proxy"] == len(keeplist.Items)
                if isExternal {
-                       diag.debugf("controller returned only proxy services, this host is treated as \"external\"")
+                       diag.verbosef("controller returned only proxy services, this host is treated as \"external\"")
                } else if isInternal {
-                       diag.debugf("controller returned only non-proxy services, this host is treated as \"internal\"")
+                       diag.verbosef("controller returned only non-proxy services, this host is treated as \"internal\"")
                }
                if (diag.checkInternal && !isInternal) || (diag.checkExternal && !isExternal) {
                        return fmt.Errorf("expecting internal=%v external=%v, but found internal=%v external=%v", diag.checkInternal, diag.checkExternal, isInternal, isExternal)
@@ -356,7 +393,7 @@ func (diag *diagnoser) runtests() {
                }
                if len(grplist.Items) > 0 {
                        project = grplist.Items[0]
-                       diag.debugf("using existing project, uuid = %s", project.UUID)
+                       diag.verbosef("using existing project, uuid = %s", project.UUID)
                        return nil
                }
                diag.debugf("list groups: ok, no results")
@@ -367,7 +404,7 @@ func (diag *diagnoser) runtests() {
                if err != nil {
                        return fmt.Errorf("create project: %s", err)
                }
-               diag.debugf("created project, uuid = %s", project.UUID)
+               diag.verbosef("created project, uuid = %s", project.UUID)
                return nil
        })
 
@@ -387,7 +424,7 @@ func (diag *diagnoser) runtests() {
                if err != nil {
                        return err
                }
-               diag.debugf("ok, uuid = %s", collection.UUID)
+               diag.verbosef("ok, uuid = %s", collection.UUID)
                return nil
        })
 
@@ -657,8 +694,8 @@ func (diag *diagnoser) runtests() {
                if err != nil {
                        return err
                }
-               diag.debugf("container request uuid = %s", cr.UUID)
-               diag.debugf("container uuid = %s", cr.ContainerUUID)
+               diag.verbosef("container request uuid = %s", cr.UUID)
+               diag.verbosef("container uuid = %s", cr.ContainerUUID)
 
                timeout := 10 * time.Minute
                diag.infof("container request submitted, waiting up to %v for container to run", arvados.Duration(timeout))

diff --git a/sdk/go/health/aggregator.go b/sdk/go/health/aggregator.go
index 63e0b0d9018f26d25cb1aa5b581bb1b35d1fbc8f..6fb33dc608054611c92c73ab41c904bada7ebb6d 100644 (file)
--- a/sdk/go/health/aggregator.go
+++ b/sdk/go/health/aggregator.go
@@ -135,6 +135,7 @@ type CheckResult struct {
        Response       map[string]interface{} `json:",omitempty"`
        ResponseTime   json.Number
        ClockTime      time.Time
+       Server         string // "Server" header in http response
        Metrics
        respTime time.Duration
 }
@@ -360,6 +361,7 @@ func (agg *Aggregator) ping(target *url.URL) (result CheckResult) {
        }
        result.Health = "OK"
        result.ClockTime, _ = time.Parse(time.RFC1123, resp.Header.Get("Date"))
+       result.Server = resp.Header.Get("Server")
        return
 }