Ensure we get at least one netdump in each topic for tested EVE upgrade

If upgrade fails, the recorded netdumps will remain persisted and
available for retrieval from the older EVE version.

Signed-off-by: Milan Lenco <milan@zededa.com>

pkg/pillar/cmd/zedagent/handleconfig.go

@@ -26,6 +26,7 @@ import (
 	"github.com/lf-edge/eve/pkg/pillar/types"
 	"github.com/lf-edge/eve/pkg/pillar/utils"
 	fileutils "github.com/lf-edge/eve/pkg/pillar/utils/file"
+	"github.com/lf-edge/eve/pkg/pillar/zboot"
 	"github.com/lf-edge/eve/pkg/pillar/zedcloud"
 	uuid "github.com/satori/go.uuid"
 	"google.golang.org/protobuf/proto"
@@ -1081,14 +1082,28 @@ func isNettraceEnabled(ctx *zedagentContext) bool {
 	return true
 }
 
-// Function decides if the next call to SendOnAllIntf for /config request should be traced
-// and netdump published at the end (see libs/nettrace and pkg/pillar/netdump).
-func traceNextConfigReq(ctx *zedagentContext) bool {
+// Function decides if the next HTTP request should be traced and netdump published.
+func traceNextReq(ctx *zedagentContext, lastNetdump time.Time) bool {
 	if !isNettraceEnabled(ctx) {
 		return false
 	}
-	return ctx.lastConfigNetdumpPub.IsZero() ||
-		time.Since(ctx.lastConfigNetdumpPub) >= ctx.netdumpInterval
+	if lastNetdump.IsZero() {
+		// No netdump published yet.
+		return true
+	}
+	uptime := time.Since(ctx.startTime)
+	lastNetdumpAge := time.Since(lastNetdump)
+	// Ensure we get at least one netdump for the currently tested EVE upgrade.
+	if zboot.IsCurrentPartitionStateInProgress() && lastNetdumpAge > uptime {
+		return true
+	}
+	return lastNetdumpAge >= ctx.netdumpInterval
+}
+
+// Function decides if the next call to SendOnAllIntf for /config request should be traced
+// and netdump published at the end (see libs/nettrace and pkg/pillar/netdump).
+func traceNextConfigReq(ctx *zedagentContext) bool {
+	return traceNextReq(ctx, ctx.lastConfigNetdumpPub)
 }
 
 // Publish netdump containing traces of executed config requests.

pkg/pillar/cmd/zedagent/reportinfo.go

@@ -1147,11 +1147,7 @@ func isUpdating(ctx *zedagentContext) bool {
 // Function decides if the next call to SendOnAllIntf for /info request should be traced
 // and netdump published at the end (see libs/nettrace and pkg/pillar/netdump).
 func traceNextInfoReq(ctx *zedagentContext) bool {
-	if !isNettraceEnabled(ctx) {
-		return false
-	}
-	return ctx.lastInfoNetdumpPub.IsZero() ||
-		time.Since(ctx.lastInfoNetdumpPub) >= ctx.netdumpInterval
+	return traceNextReq(ctx, ctx.lastInfoNetdumpPub)
 }
 
 // Publish netdump containing traces of executed /info requests.

pkg/pillar/cmd/zedagent/zedagent.go

@@ -219,6 +219,7 @@ type zedagentContext struct {
 	netdumpInterval      time.Duration
 	lastConfigNetdumpPub time.Time // last call to publishConfigNetdump
 	lastInfoNetdumpPub   time.Time // last call to publishInfoNetdump
+	startTime            time.Time
 }
 
 // AddAgentSpecificCLIFlags adds CLI options
@@ -315,6 +316,7 @@ func Run(ps *pubsub.PubSub, loggerArg *logrus.Logger, logArg *base.LogObject, ar
 	zedagentCtx.ps = ps
 	zedagentCtx.hangFlag = *zedagentCtx.hangPtr
 	zedagentCtx.fatalFlag = *zedagentCtx.fatalPtr
+	zedagentCtx.startTime = time.Now()
 
 	flowlogQueue := make(chan *flowlog.FlowMessage, flowlogQueueCap)
 	triggerDeviceInfo := make(chan destinationBitset, 1)

pkg/pillar/dpcmanager/dpcmanager.go

@@ -129,6 +129,7 @@ type DpcManager struct {
 	netDumper       *netdump.NetDumper // nil if netdump is disabled
 	netdumpInterval time.Duration
 	lastNetdumpPub  time.Time // last call to publishNetdump
+	startTime       time.Time
 }
 
 // Watchdog : methods used by DpcManager to interact with Watchdog.
@@ -238,6 +239,7 @@ func (m *DpcManager) Init(ctx context.Context) error {
 
 // Run DpcManager as a separate task with its own loop and a watchdog file.
 func (m *DpcManager) Run(ctx context.Context) (err error) {
+	m.startTime = time.Now()
 	m.networkEvents = m.NetworkMonitor.WatchEvents(ctx, "dpc-reconciler")
 	m.wwanEvents, err = m.WwanWatcher.Watch(ctx)
 	if err != nil {

pkg/pillar/dpcmanager/netdump.go

@@ -7,6 +7,7 @@ import (
 	"time"
 
 	"github.com/lf-edge/eve/pkg/pillar/netdump"
+	"github.com/lf-edge/eve/pkg/pillar/zboot"
 )
 
 // Topic for netdumps of successful connectivity tests.
@@ -31,7 +32,17 @@ func (m *DpcManager) traceNextConnTest() bool {
 	if len(m.dpcList.PortConfigList) == 0 || m.dpcList.CurrentIndex != 0 {
 		return false
 	}
-	return m.lastNetdumpPub.IsZero() || time.Since(m.lastNetdumpPub) >= m.netdumpInterval
+	if m.lastNetdumpPub.IsZero() {
+		// No netdump published yet for DPC testing.
+		return true
+	}
+	uptime := time.Since(m.startTime)
+	lastNetdumpAge := time.Since(m.lastNetdumpPub)
+	// Ensure we get at least one netdump for the currently tested EVE upgrade.
+	if zboot.IsCurrentPartitionStateInProgress() && lastNetdumpAge > uptime {
+		return true
+	}
+	return lastNetdumpAge >= m.netdumpInterval
 }
 
 // Publish netdump containing traces of executed connectivity probes.