From c75d81f4c4e347c00410dd2680859aabbc094118 Mon Sep 17 00:00:00 2001
From: David Anderson <davea@ssl.berkeley.edu>
Date: Sat, 27 Aug 2016 00:44:10 -0700
Subject: [PATCH] client: prioritize network (including GUI RPC) over async
 file ops

This is an effort to make the Manager responsive even when
lots of VM jobs are starting up.
---
 client/app_start.cpp    |  2 +-
 client/async_file.cpp   |  6 ++----
 client/async_file.h     |  5 ++++-
 client/client_state.cpp | 45 +++++++++++++++++++++--------------------
 4 files changed, 30 insertions(+), 28 deletions(-)

diff --git a/client/app_start.cpp b/client/app_start.cpp
index cc3bae52c4..b0529d9ffd 100644
--- a/client/app_start.cpp
+++ b/client/app_start.cpp
@@ -1363,7 +1363,7 @@ void run_test_app() {
         exit(1);
     }
     while (1) {
-        do_async_file_ops();
+        do_async_file_op();
         if (at.async_copy == NULL) {
             break;
         }
diff --git a/client/async_file.cpp b/client/async_file.cpp
index 7a71f128ec..e433466cb1 100644
--- a/client/async_file.cpp
+++ b/client/async_file.cpp
@@ -325,21 +325,19 @@ void remove_async_verify(ASYNC_VERIFY* avp) {
 // Note: if there are lots of pending operations,
 // it's better to finish the oldest one before starting the rest
 //
-bool do_async_file_ops() {
+void do_async_file_op() {
     if (async_copies.size()) {
         ASYNC_COPY* acp = async_copies[0];
         if (acp->copy_chunk()) {
             async_copies.erase(async_copies.begin());
             delete acp;
         }
-        return true;
+        return;
     }
     if (async_verifies.size()) {
         if (async_verifies[0]->verify_chunk()) {
             async_verifies.erase(async_verifies.begin());
         }
-        return true;
     }
-    return false;
 }
 
diff --git a/client/async_file.h b/client/async_file.h
index a86be550c3..355f2fcf30 100644
--- a/client/async_file.h
+++ b/client/async_file.h
@@ -91,6 +91,9 @@ extern std::vector<ASYNC_COPY*> async_copies;
 
 extern void remove_async_copy(ASYNC_COPY*);
 extern void remove_async_verify(ASYNC_VERIFY*);
-extern bool do_async_file_ops();
+inline bool have_async_file_op() {
+    return (async_verifies.size() || async_copies.size());
+}
+extern void do_async_file_op();
 
 #endif
diff --git a/client/client_state.cpp b/client/client_state.cpp
index 97c8697a9a..152dc99da0 100644
--- a/client/client_state.cpp
+++ b/client/client_state.cpp
@@ -809,22 +809,27 @@ FDSET_GROUP all_fds;
 
 // Spend x seconds either doing I/O (if possible) or sleeping.
 //
-void CLIENT_STATE::do_io_or_sleep(double x) {
+void CLIENT_STATE::do_io_or_sleep(double max_time) {
     int n;
     struct timeval tv;
     set_now();
-    double end_time = now + x;
-    //int loops = 0;
+    double end_time = now + max_time;
+    double time_remaining = max_time;
 
     while (1) {
-        bool action = do_async_file_ops();
-
         curl_fds.zero();
         gui_rpc_fds.zero();
         http_ops->get_fdset(curl_fds);
         all_fds = curl_fds;
         gui_rpcs.get_fdset(gui_rpc_fds, all_fds);
-        double_to_timeval(action?0:x, tv);
+
+        bool have_async = have_async_file_op();
+
+        // prioritize network (including GUI RPC) over async file ops.
+        // if there's a pending asynch file op, do the select with zero timeout;
+        // otherwise do it for the remaining amount of time.
+
+        double_to_timeval(have_async?0:time_remaining, tv);
 #ifdef NEW_CPU_THROTTLE
         client_mutex.unlock();
 #endif
@@ -843,28 +848,24 @@ void CLIENT_STATE::do_io_or_sleep(double x) {
         // called pretty often, even if no descriptors are enabled.
         // So do the "if (n==0) break" AFTER the got_selects().
 
-        http_ops->got_select(all_fds, x);
+        http_ops->got_select(all_fds, time_remaining);
         gui_rpcs.got_select(all_fds);
 
-        if (!action && n==0) break;
-
-#if 0
-        // Limit number of times thru this loop.
-        // Can get stuck in while loop, if network isn't available,
-        // DNS lookups tend to eat CPU cycles.
-        //
-        if (loops++ > 99) {
-            boinc_sleep(.01);
-#ifdef __EMX__
-            DosSleep(0);
-#endif
-            break;
+        if (have_async) {
+            // do the async file op only if no network activity
+            //
+            if (n == 0) {
+                do_async_file_op();
+            }
+        } else {
+            if (n == 0) {
+                break;
+            }
         }
-#endif
 
         set_now();
         if (now > end_time) break;
-        x = end_time - now;
+        time_remaining = end_time - now;
     }
 }