Mirror/zeek
1
0
Fork 0
mirror of https://github.com/zeek/zeek.git synced 2025-10-04 15:48:19 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

Remove timeouts from remote communication loop.

Browse source 
The select() now blocks until there's work to do instead of relying on a
small timeout value which can cause unproductive use of cpu cycles.
This commit is contained in:
Jon Siwek 2014-08-28 13:13:30 -05:00
parent 73cc81f44a
commit 675fba3fde
21 changed files with 364 additions and 58 deletions
Download patch file Download diff file Expand all files Collapse all files

47
src/RemoteSerializer.cc
View file

@ -1368,12 +1368,17 @@ void RemoteSerializer::Unregister(ID* id)
}
}
void RemoteSerializer::GetFds(int* read, int* write, int* except)
void RemoteSerializer::GetFds(std::vector<int>* read, std::vector<int>* write,
std::vector<int>* except)
{
*read = io->Fd();
read->push_back(io->Fd());
std::vector<int> supp = io->FdSupplements();
for ( size_t i = 0; i < supp.size(); ++i )
read->push_back(supp[i]);
if ( io->CanWrite() )
*write = io->Fd();
write->push_back(io->Fd());
}
double RemoteSerializer::NextTimestamp(double* local_network_time)
@ -3356,6 +3361,15 @@ SocketComm::~SocketComm()
static unsigned int first_rtime = 0;
static void fd_vector_set(const std::vector<int>& fds, fd_set* set, int* max)
{
for ( size_t i = 0; i < fds.size(); ++i )
{
FD_SET(fds[i], set);
*max = ::max(fds[i], *max);
}
}
void SocketComm::Run()
{
first_rtime = (unsigned int) current_time(true);
@ -3381,6 +3395,7 @@ void SocketComm::Run()
FD_SET(io->Fd(), &fd_read);
max_fd = io->Fd();
fd_vector_set(io->FdSupplements(), &fd_read, &max_fd);
loop_over_list(peers, i)
{
@ -3389,6 +3404,7 @@ void SocketComm::Run()
FD_SET(peers[i]->io->Fd(), &fd_read);
if ( peers[i]->io->Fd() > max_fd )
max_fd = peers[i]->io->Fd();
fd_vector_set(peers[i]->io->FdSupplements(), &fd_read, &max_fd);
}
else
{
@ -3439,38 +3455,17 @@ void SocketComm::Run()
if ( ! io->IsFillingUp() && shutting_conns_down )
shutting_conns_down = false;
// We cannot rely solely on select() as the there may
// be some data left in our input/output queues. So, we use
// a small timeout for select and check for data
// manually afterwards.
static long selects = 0;
static long canwrites = 0;
static long timeouts = 0;
++selects;
if ( io->CanWrite() )
++canwrites;
// FIXME: Fine-tune this (timeouts, flush, etc.)
struct timeval small_timeout;
small_timeout.tv_sec = 0;
small_timeout.tv_usec =
io->CanWrite() || io->CanRead() ? 1 : 10;
#if 0
if ( ! io->CanWrite() )
usleep(10);
#endif
int a = select(max_fd + 1, &fd_read, &fd_write, &fd_except,
&small_timeout);
if ( a == 0 )
++timeouts;
int a = select(max_fd + 1, &fd_read, &fd_write, &fd_except, 0);
if ( selects % 100000 == 0 )
Log(fmt("selects=%ld canwrites=%ld timeouts=%ld", selects, canwrites, timeouts));
Log(fmt("selects=%ld canwrites=%ld", selects, canwrites));
if ( a < 0 )
// Ignore errors for now.