Skip to content

Commit

Permalink
Merge pull request #6389 from STEllAR-GROUP/fixing_6388
Browse files Browse the repository at this point in the history 
Disable resolving hostnames if TCP is disabled
  • Loading branch information
@hkaiser
hkaiser authored Nov 22, 2023
2 parents d165ef8 + b9ba56a commit 5ffdfbe
Show file tree
Hide file tree
Showing 3 changed files with 60 additions and 31 deletions.
17 changes: 10 additions & 7 deletions libs/core/batch_environments/include/hpx/batch_environments/batch_environment.hpp
View file
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// Copyright (c) 2007-2022 Hartmut Kaiser
// Copyright (c) 2007-2023 Hartmut Kaiser
//
// SPDX-License-Identifier: BSL-1.0
// Distributed under the Boost Software License, Version 1.0. (See accompanying
Expand Down Expand Up @@ -31,13 +31,13 @@ namespace hpx::util {
{
// the constructor tries to read initial values from a batch
// environment, filling our map of nodes and thread counts
batch_environment(std::vector<std::string>& nodelist,
explicit batch_environment(std::vector<std::string>& nodelist,
bool have_mpi = false, bool debug = false, bool enable = true);

// this function initializes the map of nodes from the given (space
// separated) list of nodes
std::string init_from_nodelist(std::vector<std::string> const& nodes,
std::string const& agas_host);
std::string const& agas_host, bool have_tcp);

// The number of threads is either one (if no PBS information was
// found), or it is the same as the number of times this node has been
Expand Down Expand Up @@ -71,17 +71,20 @@ namespace hpx::util {
// Return a string containing the name of the batch system
std::string get_batch_name() const;

using node_map_type = std::map<asio::ip::tcp::endpoint,
std::pair<std::string, std::size_t>>;

std::string agas_node_;
std::size_t agas_node_num_;
std::size_t node_num_;
std::size_t num_threads_;
node_map_type nodes_;
std::size_t num_localities_;
std::string batch_name_;
bool debug_;

#if defined(HPX_HAVE_PARCELPORT_TCP)
using node_map_type = std::map<asio::ip::tcp::endpoint,
std::pair<std::string, std::size_t>>;

node_map_type nodes_;
#endif
};
} // namespace hpx::util

Expand Down
53 changes: 36 additions & 17 deletions libs/core/batch_environments/src/batch_environment.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// Copyright (c) 2007-2022 Hartmut Kaiser
// Copyright (c) 2007-2023 Hartmut Kaiser
// Copyright (c) 2013 Thomas Heller
//
// SPDX-License-Identifier: BSL-1.0
Expand Down Expand Up @@ -120,7 +120,8 @@ namespace hpx::util {

// this function initializes the map of nodes from the given a list of nodes
std::string batch_environment::init_from_nodelist(
std::vector<std::string> const& nodes, std::string const& agas_host)
std::vector<std::string> const& nodes, std::string const& agas_host,
bool have_tcp)
{
if (debug_)
std::cerr << "got node list" << std::endl;
Expand All @@ -139,29 +140,41 @@ namespace hpx::util {
if (debug_)
std::cerr << "extracted: '" << s << "'" << std::endl;

asio::ip::tcp::endpoint ep =
util::resolve_hostname(s, 0, io_service);

#if defined(HPX_HAVE_PARCELPORT_TCP)
if (!found_agas_host &&
((agas_host.empty() && nodes_.empty()) || s == agas_host))
{
agas_node_ = s;
found_agas_host = true;
agas_node_ = s;
agas_node_num_ = agas_node_num;
}

if (0 == nodes_.count(ep))
if (have_tcp)
{
if (debug_)
std::cerr << "incrementing agas_node_num" << std::endl;
++agas_node_num;
asio::ip::tcp::endpoint ep =
util::resolve_hostname(s, 0, io_service);

if (0 == nodes_.count(ep))
{
if (debug_)
std::cerr << "incrementing agas_node_num"
<< std::endl;
++agas_node_num;
}

std::pair<std::string, std::size_t>& data = nodes_[ep];
if (data.first.empty())
data.first = s;
++data.second;
}

std::pair<std::string, std::size_t>& data = nodes_[ep];
if (data.first.empty())
data.first = s;
++data.second;

#else
if (!found_agas_host && (agas_host.empty() || s == agas_host))
{
found_agas_host = true;
agas_node_ = s;
agas_node_num_ = agas_node_num;
}
#endif
nodes_list += s + ' ';
}
}
Expand All @@ -184,6 +197,7 @@ namespace hpx::util {
<< std::endl;
}

#if defined(HPX_HAVE_PARCELPORT_TCP)
std::cerr << "Nodes from nodelist:" << std::endl;
node_map_type::const_iterator const end = nodes_.end();
for (node_map_type::const_iterator it = nodes_.begin(); it != end;
Expand All @@ -192,14 +206,15 @@ namespace hpx::util {
std::cerr << (*it).second.first << ": " << (*it).second.second
<< " (" << (*it).first << ")" << std::endl;
}
#endif
}
HPX_UNUSED(nodes);
return nodes_list;
}

// The number of threads is either one (if no PBS/SLURM information was
// found), or it is the same as the number of times this node has been
// listed in the node file. Additionally this takes into account the number
// listed in the node file. Additionally, this takes into account the number
// of tasks run on this node.
std::size_t batch_environment::retrieve_number_of_threads() const noexcept
{
Expand Down Expand Up @@ -233,7 +248,11 @@ namespace hpx::util {
std::string batch_environment::host_name(
std::string const& def_hpx_name) const
{
#if defined(HPX_HAVE_PARCELPORT_TCP)
std::string host = nodes_.empty() ? def_hpx_name : host_name();
#else
std::string host = host_name();
#endif
if (debug_)
std::cerr << "host_name: " << host << std::endl;
return host;
Expand Down
21 changes: 14 additions & 7 deletions libs/full/command_line_handling/src/command_line_handling.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -383,6 +383,8 @@ namespace hpx::util {
node != static_cast<std::size_t>(-1) && vm.count("hpx:debug-clp");

// create host name mapping
[[maybe_unused]] bool const have_tcp =
rtcfg_.get_entry("hpx.parcel.tcp.enable", "1") != "0";
util::map_hostnames mapnames(debug_clp);

if (vm.count("hpx:ifsuffix"))
Expand Down Expand Up @@ -493,8 +495,8 @@ namespace hpx::util {
if (!nodelist.empty())
{
using_nodelist = true;
ini_config.emplace_back(
"hpx.nodes!=" + env.init_from_nodelist(nodelist, agas_host));
ini_config.emplace_back("hpx.nodes!=" +
env.init_from_nodelist(nodelist, agas_host, have_tcp));
}

// let the batch environment decide about the AGAS host
Expand All @@ -503,8 +505,8 @@ namespace hpx::util {
#endif

bool run_agas_server = false;
std::string hpx_host;
std::uint16_t hpx_port = 0;
[[maybe_unused]] std::string hpx_host;
[[maybe_unused]] std::uint16_t hpx_port = 0;

#if defined(HPX_HAVE_NETWORKING)
bool expect_connections = false;
Expand Down Expand Up @@ -724,9 +726,14 @@ namespace hpx::util {
// handle high-priority threads
handle_high_priority_threads(vm, ini_config);

#if defined(HPX_HAVE_PARCELPORT_TCP)
// map host names to ip addresses, if requested
hpx_host = mapnames.map(hpx_host, hpx_port);
agas_host = mapnames.map(agas_host, agas_port);
if (have_tcp)
{
hpx_host = mapnames.map(hpx_host, hpx_port);
agas_host = mapnames.map(agas_host, agas_port);
}
#endif

// sanity checks
if (rtcfg_.mode_ != hpx::runtime_mode::local && num_localities_ == 1 &&
Expand Down Expand Up @@ -762,7 +769,7 @@ namespace hpx::util {
}
else if (env.found_batch_environment())
{
// in batch mode, if the network addresses are different and we
// in batch mode, if the network addresses are different, and we
// should not run the AGAS server we assume to be in worker mode
rtcfg_.mode_ = hpx::runtime_mode::worker;

Expand Down

0 comments on commit 5ffdfbe

Please sign in to comment.