Reuse networks (#200)

packages/api/internal/cache/builds/build_cache.go

@@ -12,7 +12,7 @@ import (
 	"go.opentelemetry.io/otel/metric"
 
 	"github.com/e2b-dev/infra/packages/api/internal/api"
-	"github.com/e2b-dev/infra/packages/api/internal/meters"
+	"github.com/e2b-dev/infra/packages/shared/pkg/meters"
 )
 
 const (

packages/api/internal/cache/instance/instance.go

@@ -3,7 +3,6 @@ package instance
 import (
 	"context"
 	"fmt"
-	"github.com/e2b-dev/infra/packages/api/internal/meters"
 	"go.opentelemetry.io/otel/metric"
 	"sync"
 	"time"
@@ -16,6 +15,7 @@ import (
 	analyticscollector "github.com/e2b-dev/infra/packages/api/internal/analytics_collector"
 	"github.com/e2b-dev/infra/packages/api/internal/api"
 	"github.com/e2b-dev/infra/packages/shared/pkg/logs"
+	"github.com/e2b-dev/infra/packages/shared/pkg/meters"
 	"github.com/e2b-dev/infra/packages/shared/pkg/telemetry"
 )
 

packages/api/internal/handlers/sandbox_create.go

@@ -15,10 +15,10 @@ import (
 	"github.com/e2b-dev/infra/packages/api/internal/auth"
 	authcache "github.com/e2b-dev/infra/packages/api/internal/cache/auth"
 	"github.com/e2b-dev/infra/packages/api/internal/cache/instance"
-	"github.com/e2b-dev/infra/packages/api/internal/meters"
 	"github.com/e2b-dev/infra/packages/api/internal/utils"
 	"github.com/e2b-dev/infra/packages/shared/pkg/id"
 	"github.com/e2b-dev/infra/packages/shared/pkg/logs"
+	"github.com/e2b-dev/infra/packages/shared/pkg/meters"
 	"github.com/e2b-dev/infra/packages/shared/pkg/telemetry"
 )
 
@@ -101,7 +101,7 @@ func (a *APIStore) PostSandboxes(c *gin.Context) {
 	telemetry.ReportEvent(ctx, "waiting for create sandbox parallel limit semaphore slot")
 
 	_, rateSpan := a.Tracer.Start(ctx, "rate-limit")
-	counter, err := meters.GetUpDownCounter(meters.CreateRateLimitMeterName)
+	counter, err := meters.GetUpDownCounter(meters.RateLimitCounterMeterName)
 	if err != nil {
 		a.logger.Errorf("error getting counter: %s", err)
 	}

packages/nomad/otel-collector.hcl

@@ -179,13 +179,20 @@ processors:
     timeout: 5s
   filter:
     metrics:
-      exclude:
+      include:
         match_type: regexp
         # Exclude metrics that start with `http`, `go`, `rpc`, or `nomad` but aren't `nomad.client`
         metric_names:
-          - "^http.*"
-          - "^go.*"
-          - "^rpc.*"
+          - "nomad_client_host_cpu_idle"
+          - "nomad_client_host_disk_available"
+          - "nomad_client_host_disk_size"
+          - "nomad_client_allocs_memory_usage"
+          - "nomad_client_allocs_cpu_usage"
+          - "nomad_client_host_memory_available"
+          - "nomad_client_host_memory_total"
+          - "nomad_client_unallocated_memory"
+          - "orchestrator.*"
+          - "api.*"
   attributes/session-proxy:
     actions:
       - key: service.name

packages/orchestrator/internal/sandbox/cache/mmapfile.go

@@ -105,7 +105,7 @@ func NewMmapfileCache() *MmapfileCache {
 	cache.OnEviction(func(ctx context.Context, reason ttlcache.EvictionReason, item *ttlcache.Item[string, *Mmapfile]) {
 		err := item.Value().Close()
 		if err != nil {
-			fmt.Printf("failed to close mmapfile: %v", err)
+			fmt.Fprintf(os.Stderr, "failed to close mmapfile: %v", err)
 		}
 	})
 

packages/orchestrator/internal/sandbox/checks.go

@@ -42,7 +42,7 @@ func (s *Sandbox) Healthcheck(ctx context.Context, alwaysReport bool) {
 		s.Logger.Healthcheck(err == nil, alwaysReport)
 	}()
 
-	address := fmt.Sprintf("http://%s:%d/health", s.slot.HostSnapshotIP(), consts.DefaultEnvdServerPort)
+	address := fmt.Sprintf("http://%s:%d/health", s.slot.HostIP(), consts.DefaultEnvdServerPort)
 
 	request, err := http.NewRequestWithContext(ctx, "GET", address, nil)
 	if err != nil {

packages/orchestrator/internal/sandbox/fc.go

@@ -58,6 +58,13 @@ type fc struct {
 func (fc *fc) wait() error {
 	err := fc.cmd.Wait()
 	if err != nil {
+		var exitErr *exec.ExitError
+		if errors.As(err, &exitErr) {
+			// Check if the process was killed by a signal
+			if status, ok := exitErr.Sys().(syscall.WaitStatus); ok && status.Signaled() && status.Signal() == syscall.SIGKILL {
+				return nil
+			}
+		}
 		return fmt.Errorf("error waiting for fc process: %w", err)
 	}
 

packages/orchestrator/internal/sandbox/mock_instance.go

@@ -3,6 +3,7 @@ package sandbox
 import (
 	"context"
 	"fmt"
+	"os"
 	"time"
 
 	"go.opentelemetry.io/otel"
@@ -23,32 +24,21 @@ func MockInstance(envID, instanceID string, dns *dns.DNS, keepAlive time.Duratio
 
 	consulClient, err := consul.New(childCtx)
 
-	networkPool := make(chan IPSlot, 1)
-
 	logger := logs.NewSandboxLogger(instanceID, envID, "test-team", 2, 512, false)
 
-	select {
-	case <-ctx.Done():
-		return
-	default:
-		ips, err := NewSlot(ctx, tracer, consulClient)
-		if err != nil {
-			fmt.Printf("failed to create network: %v\n", err)
+	networkPool := NewNetworkSlotPool(1, 0)
 
-			return
+	go func() {
+		poolErr := networkPool.Start(ctx, consulClient)
+		if poolErr != nil {
+			fmt.Fprintf(os.Stderr, "network pool error: %v\n", poolErr)
 		}
 
-		err = ips.CreateNetwork(ctx, tracer)
-		if err != nil {
-			ips.Release(ctx, tracer, consulClient)
-
-			fmt.Printf("failed to create network: %v\n", err)
-
-			return
+		closeErr := networkPool.Close(consulClient)
+		if closeErr != nil {
+			fmt.Fprintf(os.Stderr, "network pool close error: %v\n", closeErr)
 		}
-
-		networkPool <- *ips
-	}
+	}()
 
 	start := time.Now()