Periodically recheck for manifest upload

There seems to be some obscure race, perhaps related to tmpfs?

3 files changed, 24 insertions, 13 deletions

diff --git a/bbot/agent.cxx b/bbot/agent.cxx
index 117840a..d71f7b4 100644
--- a/bbot/agent.cxx
+++ b/bbot/agent.cxx
@@ -219,9 +219,11 @@ bootstrap_machine (const dir_path& md,
         // Next the bootstrap process may download additional toolchain
         // archives, build things, and then upload the result manifest. So on
         // our side we serve TFTP requests while periodically checking for the
-        // manifest file.
+        // manifest file. To workaround some obscure filesystem races (the
+        // file's mtime/size is updated several seconds later; maybe tmpfs
+        // issue?), we periodically re-check.
         //
-        for (to = bootstrap_to; to != 0; tftpd.serve (to))
+        for (to = bootstrap_to; to != 0; tftpd.serve (to, 2))
         {
           if (file_exists (mf))
           {
@@ -702,9 +704,11 @@ try
 
         // Next the worker builds things and then uploads the result manifest.
         // So on our side we serve TFTP requests while checking for the
-        // manifest file.
+        // manifest file. To workaround some obscure filesystem races (the
+        // file's mtime/size is updated several seconds later; maybe tmpfs
+        // issue?), we periodically re-check.
         //
-        for (to = build_to; to != 0; tftpd.serve (to))
+        for (to = build_to; to != 0; tftpd.serve (to, 2))
         {
           if (file_exists (rf))
           {
diff --git a/bbot/tftp.cxx b/bbot/tftp.cxx
index 70cc85d..0da6c24 100644
--- a/bbot/tftp.cxx
+++ b/bbot/tftp.cxx
@@ -67,15 +67,18 @@ namespace bbot
   }
 
   bool tftp_server::
-  serve (size_t& sec)
+  serve (size_t& sec, size_t inc)
   {
     tracer trace ("tftp_server::serve");
 
+    if (inc == 0 || inc > sec)
+      inc = sec;
+
     int fd (fd_.get ());
 
     // Note: Linux updates the timeout value which we rely upon.
     //
-    timeval timeout {static_cast<long> (sec), 0};
+    timeval timeout {static_cast<long> (inc), 0};
 
     fd_set rd;
     FD_ZERO (&rd);
@@ -95,7 +98,7 @@ namespace bbot
       }
       else if (r == 0) // Timeout.
       {
-        sec = 0;
+        sec -= inc;
         return false;
       }
 
@@ -122,9 +125,11 @@ namespace bbot
                 "--secure",                 // Chroot to data directory.
                 ops.tftp ());
 
-        // @@ This is not really accurate.
+        // This is not really accurate since tftpd will, for example, serve
+        // an upload request until it is complete. But it's close anough for
+        // our needs.
         //
-        sec = static_cast<size_t> (timeout.tv_sec);
+        sec -= (inc - static_cast<size_t> (timeout.tv_sec));
         return true;
       }
     }
diff --git a/bbot/tftp.hxx b/bbot/tftp.hxx
index c9a88c9..cdd0e3e 100644
--- a/bbot/tftp.hxx
+++ b/bbot/tftp.hxx
@@ -29,12 +29,14 @@ namespace bbot
     uint16_t
     port () const;
 
-    // Wait for a TFTP request for up to the specified number of seconds.
-    // Update the timeout value as well as return true if a request was
-    // served and false otherwise.
+    // Wait for a TFTP request for up to the specified number of seconds.  If
+    // increment is not 0, then wait in the specified incremenets (i.e., wait
+    // for up to that number of seconds; useful when one needs to also
+    // periodically check for something else). Update the timeout value as
+    // well as return true if a request was served and false otherwise.
     //
     bool
-    serve (size_t& seconds);
+    serve (size_t& seconds, size_t increment = 0);
 
   private:
     auto_fd fd_;