memory-infra: Don't consider clients with null PIDs

After crrev.com/c/565510 the PID is received asynchronously
after the start of a service. It is possible to get a dump
request between the point when a client process is started
and the point when the PID is received. In this case just
temporarily skip that client's dumps.

Bug: 744722
Change-Id: I69a70747408232063a1d57933d8d0ff985e153c9
Reviewed-on: https://chromium-review.googlesource.com/575127
Commit-Queue: Hector Dearman <hjd@chromium.org>
Reviewed-by: Primiano Tucci <primiano@chromium.org>
Cr-Commit-Position: refs/heads/master@{#487487}

services/resource_coordinator/memory_instrumentation/coordinator_impl.cc

@@ -238,6 +238,7 @@ void CoordinatorImpl::PerformNextQueuedGlobalMemoryDump() {
       // TODO(hjd): Change to NOTREACHED when crbug.com/739710 is fixed.
       VLOG(1) << "Couldn't find a PID for client \"" << client_identity.name()
               << "." << client_identity.instance() << "\"";
+      continue;
     }
     request->responses[client].process_id = pid;
     request->responses[client].process_type = kv.second->process_type;
@@ -265,6 +266,12 @@ void CoordinatorImpl::PerformNextQueuedGlobalMemoryDump() {
   for (const auto& kv : clients_) {
     service_manager::Identity client_identity = kv.second->identity;
     const base::ProcessId pid = GetProcessIdForClientIdentity(client_identity);
+    if (pid == base::kNullProcessId) {
+      // TODO(hjd): Change to NOTREACHED when crbug.com/739710 is fixed.
+      VLOG(1) << "Couldn't find a PID for client \"" << client_identity.name()
+              << "." << client_identity.instance() << "\"";
+      continue;
+    }
     pids.push_back(pid);
     if (kv.second->process_type == mojom::ProcessType::BROWSER) {
       browser_client = kv.first;
@@ -416,10 +423,8 @@ void CoordinatorImpl::FinalizeGlobalMemoryDumpIfAllManagersReplied() {
   std::map<base::ProcessId, mojom::ProcessMemoryDumpPtr> finalized_pmds;
   for (auto& response : request->responses) {
     const base::ProcessId pid = response.second.process_id;
-    // TODO(hjd): this shouldn't really happen, but seem to be hitting this in
-    // crbug.com/744722 . Temporarily softening the DCHECK while investigating.
-    if (finalized_pmds.count(pid) > 0)
-      continue;
+    DCHECK(!finalized_pmds.count(pid))
+        << "Received PID: " << pid << " more than once.";
 
     // The dump might be nullptr if the client crashed / disconnected before
     // replying.