Add --health-max-log-count flag

Signed-off-by: Jan Rodák <[email protected]>

cmd/podman/common/create.go

@@ -184,6 +184,14 @@ func DefineCreateFlags(cmd *cobra.Command, cf *entities.ContainerCreateOptions,
 		)
 		_ = cmd.RegisterFlagCompletionFunc(healthIntervalFlagName, completion.AutocompleteNone)
 
+		healthMaxLogCountFlagName := "health-max-log-count"
+		createFlags.UintVar(
+			&cf.HealthMaxLogCount,
+			healthMaxLogCountFlagName, define.DefaultHealthMaxLogCount,
+			"set the maximum number of attempts we keep in the healthcheck history file. ('0' value means no limit for stored logs)",
+		)
+		_ = cmd.RegisterFlagCompletionFunc(healthMaxLogCountFlagName, completion.AutocompleteNone)
+
 		healthRetriesFlagName := "health-retries"
 		createFlags.UintVar(
 			&cf.HealthRetries,

docs/source/markdown/options/health-max-log-count.md

@@ -0,0 +1,7 @@
+####> This option file is used in:
+####>   podman create, run
+####> If file is edited, make sure the changes
+####> are applicable to all of those.
+#### **--health-max-log-count**=*number of stored logs*
+
+Set the maximum number of attempts we keep in the healthcheck history file. ('0' value means no limit for stored logs)

docs/source/markdown/podman-create.1.md.in

@@ -169,6 +169,8 @@ See [**Environment**](#environment) note below for precedence and examples.
 
 @@option health-interval
 
+@@option health-max-log-count
+
 @@option health-on-failure
 
 @@option health-retries

docs/source/markdown/podman-run.1.md.in

@@ -203,6 +203,8 @@ See [**Environment**](#environment) note below for precedence and examples.
 
 @@option health-interval
 
+@@option health-max-log-count
+
 @@option health-on-failure
 
 @@option health-retries

docs/source/markdown/podman-systemd.unit.5.md

@@ -279,6 +279,7 @@ Valid options for `[Container]` are listed below:
 | GroupAdd=keep-groups                 | --group-add=keep-groups                              |
 | HealthCmd=/usr/bin/command           | --health-cmd=/usr/bin/command                        |
 | HealthInterval=2m                    | --health-interval=2m                                 |
+| HealthMaxLogCount=5                  | --health-max-log-count=5                             |
 | HealthOnFailure=kill                 | --health-on-failure=kill                             |
 | HealthRetries=5                      | --health-retries=5                                   |
 | HealthStartPeriod=1m                 | --health-start-period=period=1m                      |
@@ -515,6 +516,11 @@ Equivalent to the Podman `--health-cmd` option.
 Set an interval for the healthchecks. An interval of disable results in no automatic timer setup.
 Equivalent to the Podman `--health-interval` option.
 
+### `HealthMaxLogCount=`
+
+Set the maximum number of attempts we keep in the healthcheck history file. ('0' value means no limit for stored logs)
+Equivalent to the Podman `--Health-max-log-count` option.
+
 ### `HealthOnFailure=`
 
 Action to take once the container transitions to an unhealthy state.

libpod/container_config.go

@@ -413,6 +413,9 @@ type ContainerMiscConfig struct {
 	HealthCheckConfig *manifest.Schema2HealthConfig `json:"healthcheck"`
 	// HealthCheckOnFailureAction defines an action to take once the container turns unhealthy.
 	HealthCheckOnFailureAction define.HealthCheckOnFailureAction `json:"healthcheck_on_failure_action"`
+	// HealthMaxLogCount is the maximum number of attempts we keep
+	// in the healthcheck history file ("0" value means no limit)
+	HealthMaxLogCount uint `json:"healthMaxLogCount"`
 	// StartupHealthCheckConfig is the configuration of the startup
 	// healthcheck for the container. This will run before the regular HC
 	// runs, and when it passes the regular HC will be activated.

libpod/container_inspect.go

@@ -412,6 +412,8 @@ func (c *Container) generateInspectContainerConfig(spec *spec.Spec) *define.Insp
 
 	ctrConfig.HealthcheckOnFailureAction = c.config.HealthCheckOnFailureAction.String()
 
+	ctrConfig.HealthMaxLogCount = c.config.HealthMaxLogCount
+
 	ctrConfig.CreateCommand = c.config.CreateCommand
 
 	ctrConfig.Timezone = c.config.Timezone

libpod/define/container_inspect.go

@@ -61,6 +61,9 @@ type InspectContainerConfig struct {
 	Healthcheck *manifest.Schema2HealthConfig `json:"Healthcheck,omitempty"`
 	// HealthcheckOnFailureAction defines an action to take once the container turns unhealthy.
 	HealthcheckOnFailureAction string `json:"HealthcheckOnFailureAction,omitempty"`
+	// HealthMaxLogCount is the maximum number of attempts we keep
+	// in the healthcheck history file ("0" value means no limit)
+	HealthMaxLogCount uint `json:"HealthcheckMaxLogCount,omitempty"`
 	// CreateCommand is the full command plus arguments of the process the
 	// container has been created with.
 	CreateCommand []string `json:"CreateCommand,omitempty"`

libpod/define/healthchecks.go

@@ -56,6 +56,8 @@ const (
 	DefaultHealthCheckStartPeriod = "0s"
 	// DefaultHealthCheckTimeout default value
 	DefaultHealthCheckTimeout = "30s"
+	// DefaultHealthMaxLogCount default value
+	DefaultHealthMaxLogCount uint = 5
 )
 
 // HealthConfig.Test options

libpod/healthcheck.go

@@ -20,9 +20,6 @@ import (
 )
 
 const (
-	// MaxHealthCheckNumberLogs is the maximum number of attempts we keep
-	// in the healthcheck history file
-	MaxHealthCheckNumberLogs int = 5
 	// MaxHealthCheckLogLength in characters
 	MaxHealthCheckLogLength = 500
 )
@@ -398,7 +395,7 @@ func (c *Container) updateHealthCheckLog(hcl define.HealthCheckLog, inStartPerio
 		}
 	}
 	healthCheck.Log = append(healthCheck.Log, hcl)
-	if len(healthCheck.Log) > MaxHealthCheckNumberLogs {
+	if c.config.HealthMaxLogCount != 0 && len(healthCheck.Log) > int(c.config.HealthMaxLogCount) {
 		healthCheck.Log = healthCheck.Log[1:]
 	}
 	newResults, err := json.Marshal(healthCheck)

libpod/options.go

@@ -1500,6 +1500,17 @@ func WithHealthCheck(healthCheck *manifest.Schema2HealthConfig) CtrCreateOption
 	}
 }
 
+// WithHealthCheckMaxLogCount adds the healthCheckMaxLogCount to the container config
+func WithHealthCheckMaxLogCount(maxLogCount uint) CtrCreateOption {
+	return func(ctr *Container) error {
+		if ctr.valid {
+			return define.ErrCtrFinalized
+		}
+		ctr.config.HealthMaxLogCount = maxLogCount
+		return nil
+	}
+}
+
 // WithHealthCheckOnFailureAction adds an on-failure action to health-check config
 func WithHealthCheckOnFailureAction(action define.HealthCheckOnFailureAction) CtrCreateOption {
 	return func(ctr *Container) error {

pkg/domain/entities/pods.go

@@ -171,6 +171,7 @@ type ContainerCreateOptions struct {
 	HealthCmd          string
 	HealthInterval     string
 	HealthRetries      uint
+	HealthMaxLogCount  uint
 	HealthStartPeriod  string
 	HealthTimeout      string
 	HealthOnFailure    string

pkg/specgen/generate/container_create.go

@@ -642,6 +642,10 @@ func createContainerOptions(rt *libpod.Runtime, s *specgen.SpecGenerator, pod *l
 		options = append(options, libpod.WithHealthCheckOnFailureAction(s.ContainerHealthCheckConfig.HealthCheckOnFailureAction))
 	}
 
+	if healthCheckSet {
+		options = append(options, libpod.WithHealthCheckMaxLogCount(s.ContainerHealthCheckConfig.HealthMaxLogCount))
+	}
+
 	if s.SdNotifyMode == define.SdNotifyModeHealthy && !healthCheckSet {
 		return nil, fmt.Errorf("%w: sdnotify policy %q requires a healthcheck to be set", define.ErrInvalidArg, s.SdNotifyMode)
 	}

pkg/specgen/specgen.go

@@ -599,6 +599,9 @@ type ContainerHealthCheckConfig struct {
 	// Requires that HealthConfig be set.
 	// Optional.
 	StartupHealthConfig *define.StartupHealthCheck `json:"startupHealthConfig,omitempty"`
+	// HealthMaxLogCount is the maximum number of attempts we keep
+	// in the healthcheck history file ("0" value means no limit)
+	HealthMaxLogCount uint `json:"healthMaxLogCount,omitempty"`
 }
 
 // SpecGenerator creates an OCI spec and Libpod configuration options to create

pkg/specgenutil/specgen.go

@@ -370,6 +370,8 @@ func FillOutSpecGen(s *specgen.SpecGenerator, c *entities.ContainerCreateOptions
 	}
 	s.HealthCheckOnFailureAction = onFailureAction
 
+	s.HealthMaxLogCount = c.HealthMaxLogCount
+
 	if c.StartupHCCmd != "" {
 		if c.NoHealthCheck {
 			return errors.New("cannot specify both --no-healthcheck and --health-startup-cmd")

pkg/systemd/quadlet/quadlet.go

@@ -93,6 +93,7 @@ const (
 	KeyGroupAdd              = "GroupAdd"
 	KeyHealthCmd             = "HealthCmd"
 	KeyHealthInterval        = "HealthInterval"
+	KeyHealthMaxLogCount     = "HealthMaxLogCount"
 	KeyHealthOnFailure       = "HealthOnFailure"
 	KeyHealthRetries         = "HealthRetries"
 	KeyHealthStartPeriod     = "HealthStartPeriod"
@@ -214,6 +215,7 @@ var (
 		KeyHealthCmd:             true,
 		KeyHealthInterval:        true,
 		KeyHealthOnFailure:       true,
+		KeyHealthMaxLogCount:     true,
 		KeyHealthRetries:         true,
 		KeyHealthStartPeriod:     true,
 		KeyHealthStartupCmd:      true,
@@ -2065,6 +2067,7 @@ func handleHealth(unitFile *parser.UnitFile, groupName string, podman *PodmanCmd
 		{KeyHealthCmd, "cmd"},
 		{KeyHealthInterval, "interval"},
 		{KeyHealthOnFailure, "on-failure"},
+		{KeyHealthMaxLogCount, "max-log-count"},
 		{KeyHealthRetries, "retries"},
 		{KeyHealthStartPeriod, "start-period"},
 		{KeyHealthTimeout, "timeout"},

test/e2e/healthcheck_run_test.go

@@ -390,4 +390,56 @@ HEALTHCHECK CMD ls -l / 2>&1`, ALPINE)
 		Expect(ps.OutputToStringArray()).To(HaveLen(2))
 		Expect(ps.OutputToString()).To(ContainSubstring("hc"))
 	})
+
+	It("Healthcheck with max default value (5) of last execution log", func() {
+		countOfExecutions := 10
+		ctrName := "hc"
+		ctrRun := podmanTest.Podman([]string{"run", "-dt", "--name", ctrName, "--health-cmd", "echo hello", ALPINE, "top"})
+		ctrRun.WaitWithDefaultTimeout()
+		Expect(ctrRun).Should(ExitCleanly())
+
+		for i := 0; i < countOfExecutions; i++ {
+			hc := podmanTest.Podman([]string{"healthcheck", "run", ctrName})
+			hc.WaitWithDefaultTimeout()
+			Expect(hc).Should(ExitCleanly())
+		}
+
+		inspect := podmanTest.InspectContainer(ctrName)
+		Expect(inspect[0].State.Health.Log).To(HaveLen(5))
+	})
+
+	It("Healthcheck with max infinite value (0) of last execution log", func() {
+		countOfExecutions := 12
+		ctrName := "hc"
+		ctrRun := podmanTest.Podman([]string{"run", "-dt", "--name", ctrName, "--health-cmd", "echo hello", "--health-max-log-count", "0", ALPINE, "top"})
+		ctrRun.WaitWithDefaultTimeout()
+		Expect(ctrRun).Should(ExitCleanly())
+
+		for i := 0; i < countOfExecutions; i++ {
+			hc := podmanTest.Podman([]string{"healthcheck", "run", ctrName})
+			hc.WaitWithDefaultTimeout()
+			Expect(hc).Should(ExitCleanly())
+		}
+
+		inspect := podmanTest.InspectContainer(ctrName)
+		Expect(inspect[0].State.Health.Log).To(HaveLen(countOfExecutions))
+	})
+
+	It("Healthcheck with max 10 last execution log", func() {
+		countOfExecutions := 10
+		ctrName := "hc"
+		ctrRun := podmanTest.Podman([]string{"run", "-dt", "--name", ctrName, "--health-cmd", "echo hello", "--health-max-log-count", strconv.Itoa(countOfExecutions), ALPINE, "top"})
+		ctrRun.WaitWithDefaultTimeout()
+		Expect(ctrRun).Should(ExitCleanly())
+
+		for i := 0; i < countOfExecutions; i++ {
+			hc := podmanTest.Podman([]string{"healthcheck", "run", ctrName})
+			hc.WaitWithDefaultTimeout()
+			Expect(hc).Should(ExitCleanly())
+		}
+
+		inspect := podmanTest.InspectContainer(ctrName)
+		Expect(inspect[0].State.Health.Log).To(HaveLen(countOfExecutions))
+	})
+
 })

test/system/220-healthcheck.bats

@@ -268,4 +268,84 @@ Log[-1].Output   | \"Uh-oh on stdout!\\\nUh-oh on stderr!\\\n\"
     done
 }
 
+@test "podman healthcheck --health-max-log-count default value (5)" {
+    local repeat_count=10
+    local msg="Hello, How are you?"
+    local ctrname="c-h-$(safename)"
+    run_podman run -d --name $ctrname   \
+               --health-cmd "echo $msg" \
+               $IMAGE /home/podman/pause
+    cid="$output"
+
+    run_podman inspect $ctrname --format "{{.Config.HealthMaxLogCount}}"
+    is "$output" "5" "HealthMaxLogCount is set to 5"
+
+    for _ in $(seq 1 $repeat_count);
+    do
+        run_podman healthcheck run $ctrname
+        is "$output" "" "output from 'podman healthcheck run'"
+    done
+
+
+    run_podman inspect $ctrname --format "{{.State.Health.Log}}"
+    [ $(echo "$output" | grep -o "$msg" - | wc -l) -eq 5 ]
+
+    run_podman rm -t 0 -f $ctrname
+}
+
+@test "podman healthcheck --health-max-log-count infinite value (0)" {
+    local repeat_count=10
+    local msg="Hello, How are you?"
+    local ctrname="c-h-$(safename)"
+    run_podman run -d --name $ctrname   \
+               --health-cmd "echo $msg" \
+               --health-max-log-count 0 \
+               $IMAGE /home/podman/pause
+    cid="$output"
+
+    run_podman inspect $ctrname --format "{{.Config.HealthMaxLogCount}}"
+    is "$output" "0" "HealthMaxLogCount is set to 0"
+
+    # This is run 11 times to check that the cap is working.
+    for _ in $(seq 0 $repeat_count);
+    do
+        run_podman healthcheck run $ctrname
+        is "$output" "" "output from 'podman healthcheck run'"
+    done
+
+
+    run_podman inspect $ctrname --format "{{.State.Health.Log}}"
+    [ $(echo "$output" | grep -o "$msg" - | wc -l) -eq 11 ]
+
+    run_podman rm -t 0 -f $ctrname
+}
+
+
+@test "podman healthcheck --health-max-log-count 10" {
+    local repeat_count=10
+    local msg="Hello, How are you?"
+    local ctrname="c-h-$(safename)"
+    run_podman run -d --name $ctrname   \
+               --health-cmd "echo $msg" \
+               --health-max-log-count $repeat_count\
+               $IMAGE /home/podman/pause
+    cid="$output"
+
+    run_podman inspect $ctrname --format "{{.Config.HealthMaxLogCount}}"
+    is "$output" "10" "HealthMaxLogCount is set to 10"
+
+    # This is run 11 times to check that the cap is working.
+    for _ in $(seq 0 $repeat_count);
+    do
+        run_podman healthcheck run $ctrname
+        is "$output" "" "output from 'podman healthcheck run'"
+    done
+
+
+    run_podman inspect $ctrname --format "{{.State.Health.Log}}"
+    [ $(echo "$output" | grep -o "$msg" - | wc -l) -eq $repeat_count ]
+
+    run_podman rm -t 0 -f $ctrname
+}
+
 # vim: filetype=sh