Merge branch 'master' of https://github.com/buildbuddy-io/buildbuddy

VERSION

@@ -1 +1 @@
-v2.10.3
+v2.10.4

WORKSPACE

@@ -69,10 +69,9 @@ gazelle_dependencies()
 
 http_archive(
     name = "aspect_rules_js",
-    #sha256 = "ac67010f1c150a70d7ebf0026754b8a754f3be6d899b699f97be3c046a951c6a",
-    # ALMOST 1.0.0
-    strip_prefix = "rules_js-6a2a3477fcbcd819cd7f44d5bfd47f2fed30add4",
-    url = "https://github.com/aspect-build/rules_js/archive/6a2a3477fcbcd819cd7f44d5bfd47f2fed30add4.tar.gz",
+    sha256 = "e7a21d8eec5565da01417bc809213405271a818e870890c2d3bb3dde5e808bcc",
+    strip_prefix = "rules_js-1.0.0-beta.0",
+    url = "https://github.com/aspect-build/rules_js/archive/refs/tags/v1.0.0-beta.0.tar.gz",
 )
 
 load("@aspect_rules_js//js:repositories.bzl", "rules_js_dependencies")
@@ -310,9 +309,9 @@ swc_register_toolchains(
 
 http_archive(
     name = "aspect_rules_ts",
-    #sha256 = "b79eca71668f1d5e318c25a25a5f9a150d351cbfab1ea9e225b0ee7a9f16763d",
-    strip_prefix = "rules_ts-d4a70a7389a51dbd2c1a2a06c99c38e6acd4e7e0",
-    url = "https://github.com/aspect-build/rules_ts/archive/d4a70a7389a51dbd2c1a2a06c99c38e6acd4e7e0.tar.gz",
+    sha256 = "991a5ccad5fd276164ea64c01ae0b67820a5d514fbf37ae3f7ac8701a84b9f5a",
+    strip_prefix = "rules_ts-0.7.0",
+    url = "https://github.com/aspect-build/rules_ts/archive/refs/tags/v0.7.0.tar.gz",
 )
 
 load("@aspect_rules_ts//ts:repositories.bzl", "rules_ts_dependencies")

enterprise/server/scheduling/scheduler_client/scheduler_client.go

@@ -109,14 +109,11 @@ func (r *Registration) Check(ctx context.Context) error {
 	return errors.New("not registered to scheduler yet")
 }
 
-func (r *Registration) processWorkStream(ctx context.Context, stream scpb.Scheduler_RegisterAndStreamWorkClient, schedulerMsgs chan *scpb.RegisterAndStreamWorkResponse) (bool, error) {
+func (r *Registration) processWorkStream(ctx context.Context, stream scpb.Scheduler_RegisterAndStreamWorkClient, schedulerMsgs chan *scpb.RegisterAndStreamWorkResponse, registrationTicker *time.Ticker) (bool, error) {
 	registrationMsg := &scpb.RegisterAndStreamWorkRequest{
 		RegisterExecutorRequest: &scpb.RegisterExecutorRequest{Node: r.node},
 	}
 
-	idleTimer := time.NewTimer(schedulerCheckInInterval)
-	defer idleTimer.Stop()
-
 	select {
 	case <-ctx.Done():
 		log.Debugf("Context cancelled, cancelling node registration.")
@@ -157,9 +154,9 @@ func (r *Registration) processWorkStream(ctx context.Context, stream scpb.Schedu
 		if err := stream.Send(rspMsg); err != nil {
 			return false, status.UnavailableErrorf("could not send task reservation response: %s", err)
 		}
-	case <-idleTimer.C:
+	case <-registrationTicker.C:
 		if err := stream.Send(registrationMsg); err != nil {
-			return false, status.UnavailableErrorf("could not send idle registration message: %s", err)
+			return false, status.UnavailableErrorf("could not send registration message: %s", err)
 		}
 	}
 	return false, nil
@@ -207,8 +204,9 @@ func (r *Registration) maintainRegistrationAndStreamWork(ctx context.Context) {
 			}
 		}()
 
+		registrationTicker := time.NewTicker(schedulerCheckInInterval)
 		for {
-			done, err := r.processWorkStream(ctx, stream, schedulerMsgs)
+			done, err := r.processWorkStream(ctx, stream, schedulerMsgs, registrationTicker)
 			if err != nil {
 				_ = stream.CloseSend()
 				log.Warningf("Error maintaining registration with scheduler, will retry: %s", err)

enterprise/server/scheduling/scheduler_server/BUILD

@@ -29,6 +29,7 @@ go_library(
         "@org_golang_google_grpc//peer",
         "@org_golang_google_protobuf//encoding/prototext",
         "@org_golang_google_protobuf//proto",
+        "@org_golang_google_protobuf//types/known/timestamppb",
     ],
 )
 

enterprise/server/scheduling/scheduler_server/scheduler_server.go

@@ -28,6 +28,7 @@ import (
 	"google.golang.org/grpc/peer"
 	"google.golang.org/protobuf/encoding/prototext"
 	"google.golang.org/protobuf/proto"
+	"google.golang.org/protobuf/types/known/timestamppb"
 
 	remote_execution_config "github.com/buildbuddy-io/buildbuddy/enterprise/server/remote_execution/config"
 	scheduler_server_config "github.com/buildbuddy-io/buildbuddy/enterprise/server/scheduling/scheduler_server/config"
@@ -41,6 +42,7 @@ var (
 	defaultPoolName              = flag.String("remote_execution.default_pool_name", "", "The default executor pool to use if one is not specified.")
 	sharedExecutorPoolGroupID    = flag.String("remote_execution.shared_executor_pool_group_id", "", "Group ID that owns the shared executor pool.")
 	requireExecutorAuthorization = flag.Bool("remote_execution.require_executor_authorization", false, "If true, executors connecting to this server must provide a valid executor API key.")
+	removeStaleExecutors         = flag.Bool("remote_execution.remove_stale_executors", false, "If true, executors are removed if they are not heard from for a prolonged amount of time.")
 )
 
 const (
@@ -60,6 +62,10 @@ const (
 	// was fetched more than this duration ago, they will be re-fetched.
 	maxAllowedExecutionNodesStaleness = 10 * time.Second
 
+	// An executor is removed if it does not refresh its registration within
+	// this amount of time.
+	executorMaxRegistrationStaleness = 10 * time.Minute
+
 	// The maximum number of times a task may be re-enqueued.
 	maxTaskAttemptCount = 5
 
@@ -440,6 +446,15 @@ func (np *nodePool) fetchExecutionNodes(ctx context.Context) ([]*executionNode,
 		if err != nil {
 			return nil, err
 		}
+
+		if *removeStaleExecutors && time.Since(node.GetLastPingTime().AsTime()) > executorMaxRegistrationStaleness {
+			log.Infof("Removing stale executor %q from pool %+v", id, np.key)
+			if err := np.rdb.HDel(ctx, np.key.redisPoolKey(), id).Err(); err != nil {
+				log.Warningf("could not remove stale executor: %s", err)
+			}
+			continue
+		}
+
 		executors = append(executors, &executionNode{
 			executorID:            id,
 			schedulerHostPort:     node.GetSchedulerHostPort(),
@@ -900,6 +915,7 @@ func (s *SchedulerServer) insertOrUpdateNode(ctx context.Context, executorHandle
 		SchedulerHostPort: s.ownHostPort,
 		GroupId:           groupID,
 		Acl:               acl,
+		LastPingTime:      timestamppb.Now(),
 	}
 	b, err := proto.Marshal(r)
 	if err != nil {

package.json

@@ -25,26 +25,19 @@
     "@types/react-modal": "^3.10.6",
     "@types/react-virtualized": "^9.21.9",
     "@types/uuid": "^8.3.0",
-    "chalk": "^4.0.0",
     "clang-format": "^1.4.0",
     "concurrently": "^5.3.0",
     "dagre-d3-react": "^0.2.4",
     "date-fns": "^2.23.0",
     "diff": "^5.0.0",
     "diff-match-patch": "^1.0.5",
-    "escodegen": "^2.0.0",
-    "espree": "^7.0.0",
-    "estraverse": "^5.1.0",
     "events": "^3.2.0",
-    "glob": "^7.1.4",
     "google-protobuf": "^3.14.0",
     "grpc-web": "^1.2.1",
     "jasmine": "^4.1.0",
-    "jsdoc": "^3.6.2",
     "lint-staged": "^10.2.13",
     "long": "^4.0.0",
     "lucide-react": "^0.16.29",
-    "minimist": "^1.2.6",
     "moment": "^2.29.2",
     "monaco-editor": "^0.25.2",
     "octokit": "^1.1.0",
@@ -62,11 +55,8 @@
     "recharts": "^2.1.9",
     "rollup-plugin-replace": "^2.2.0",
     "rxjs": "^6.6.3",
-
-    "tmp": "^0.2.0",
     "tslib": "^2.1.0",
     "typescript": "4.7.3",
-    "uglify-js": "^3.6.0",
     "uuid": "^8.3.0"
   },
   "resolutions": {

proto/BUILD

@@ -394,6 +394,7 @@ proto_library(
         ":acl_proto",
         ":context_proto",
         ":trace_proto",
+        "@com_google_protobuf//:timestamp_proto",
     ],
 )
 

proto/defs.bzl

@@ -1,32 +1,7 @@
 load("@aspect_rules_js//js:defs.bzl", "js_library")
-
-# TODO switch to protobufjs-cli when its published
-# https://github.com/protobufjs/protobuf.js/commit/da34f43ccd51ad97017e139f137521782f5ef119
 load("@npm//protobufjs:package_json.bzl", "bin")
 load("@rules_proto//proto:defs.bzl", "ProtoInfo")
 
-# protobuf.js relies on these packages, but does not list them as dependencies
-# in its package.json.
-# Instead they are listed under "cliDependencies"
-# (see https://unpkg.com/protobufjs@6.10.2/package.json)
-# When run, the CLI attempts to run `npm install` at runtime to get them.
-# This fails under Bazel as it tries to access the npm cache outside of the sandbox.
-# Per Bazel semantics, all dependencies should be pre-declared.
-# Note, you'll also need to install all of these in your package.json!
-# (This should be fixed when we switch to protobufjs-cli)
-# NOTE: this is unused since pnpm.packageExtensions in package.json fixes this right within build graph.
-_PROTOBUFJS_CLI_DEPS = ["//:node_modules/%s" % s for s in [
-    "chalk",
-    "escodegen",
-    "espree",
-    "estraverse",
-    "glob",
-    "jsdoc",
-    "minimist",
-    "semver",
-    "tmp",
-    "uglify-js",
-]]
 
 def _proto_sources_impl(ctx):
     return DefaultInfo(files = ctx.attr.proto[ProtoInfo].transitive_sources)
@@ -40,7 +15,7 @@ _proto_sources = rule(
     attrs = {"proto": attr.label(providers = [ProtoInfo])},
 )
 
-def ts_proto_library(name, proto, **kwargs):
+def ts_proto_library(name, proto, deps = [], **kwargs):
     """Minimal wrapper macro around pbjs/pbts tooling
 
     Args:
@@ -69,7 +44,9 @@ def ts_proto_library(name, proto, **kwargs):
         name = js_target,
         srcs = [":" + proto_target],
         copy_srcs_to_bin = False,
-        chdir = "../../../",
+        env = {
+            "BAZEL_BINDIR": "."
+        },
         # Arguments documented at
         # https://github.com/protobufjs/protobuf.js/tree/6.8.8#pbjs-for-javascript
         args = [
@@ -88,7 +65,9 @@ def ts_proto_library(name, proto, **kwargs):
         name = ts_target,
         srcs = [js_target],
         copy_srcs_to_bin = False,
-        chdir = "../../../",
+        env = {
+            "BAZEL_BINDIR": "."
+        },
         # Arguments documented at
         # https://github.com/protobufjs/protobuf.js/tree/6.8.8#pbts-for-typescript
         args = [
@@ -98,21 +77,13 @@ def ts_proto_library(name, proto, **kwargs):
         outs = [ts_out],
     )
 
-    # umd_bundle(
-    #     name = name + "__umd",
-    #     package_name = name,
-    #     entry_point = ":" + js_out,
-    # )
-
     # Expose the results as js_library which provides DeclarationInfo for interop with other rules
-    if "deps" not in kwargs:
-        kwargs["deps"] = []
-    kwargs["deps"].append("//:node_modules/protobufjs")
     js_library(
         name = name,
         srcs = [
             js_target,
             ts_target,
         ],
+        deps = deps + ["//:node_modules/protobufjs"],
         **kwargs
     )

proto/scheduler.proto

@@ -1,5 +1,6 @@
 syntax = "proto3";
 
+import "google/protobuf/timestamp.proto";
 import "proto/acl.proto";
 import "proto/context.proto";
 import "proto/trace.proto";
@@ -238,4 +239,5 @@ message RegisteredExecutionNode {
   string scheduler_host_port = 2;
   string group_id = 3;
   acl.ACL acl = 4;
+  google.protobuf.Timestamp last_ping_time = 5;
 }