From ba6868963187c34af32af833ba23b29a899e224d Mon Sep 17 00:00:00 2001
From: Felix Weiglhofer <weiglhofer@fias.uni-frankfurt.de>
Date: Tue, 6 Jun 2023 14:19:11 +0000
Subject: [PATCH] algo: Add monitoring support in cbmreco.
---
algo/CMakeLists.txt | 1 +
algo/base/ChainContext.h | 4 ++++
algo/base/Options.cxx | 2 ++
algo/base/Options.h | 2 ++
algo/base/SubChain.h | 8 ++++++++
algo/base/util/TsUtils.h | 19 +++++++++++++++++++
algo/global/Reco.cxx | 11 +++++++++++
7 files changed, 47 insertions(+)
create mode 100644 algo/base/util/TsUtils.h
diff --git a/algo/CMakeLists.txt b/algo/CMakeLists.txt
index 2f8182f323..d107ef3998 100644
--- a/algo/CMakeLists.txt
+++ b/algo/CMakeLists.txt
@@ -83,6 +83,7 @@ target_link_libraries(Algo
external::yaml-cpp
external::fles_logging
external::fles_ipc
+ external::fles_monitoring
)
target_compile_definitions(Algo PUBLIC NO_ROOT)
xpu_attach(Algo ${DEVICE_SRCS})
diff --git a/algo/base/ChainContext.h b/algo/base/ChainContext.h
index 3277028b2a..0ed47dcc38 100644
--- a/algo/base/ChainContext.h
+++ b/algo/base/ChainContext.h
@@ -4,6 +4,9 @@
#ifndef CBM_ALGO_BASE_CHAINCONTEXT_H
#define CBM_ALGO_BASE_CHAINCONTEXT_H
+#include <Monitor.hpp>
+#include <optional>
+
#include "Options.h"
#include "RecoParams.h"
@@ -12,6 +15,7 @@ namespace cbm::algo
struct ChainContext {
Options opts;
RecoParams recoParams;
+ std::optional<cbm::Monitor> monitor;
};
} // namespace cbm::algo
diff --git a/algo/base/Options.cxx b/algo/base/Options.cxx
index ff74e0e559..efd141f09a 100644
--- a/algo/base/Options.cxx
+++ b/algo/base/Options.cxx
@@ -52,6 +52,8 @@ Options::Options(int argc, char** argv)
"select device (cpu, cuda0, cuda1, hip0, ...)")
("log-level,l", po::value(&fLogLevel)->default_value(info)->value_name("<level>"),
"set log level (debug, info, warning, error, fatal)")
+ ("monitor,m", po::value(&fMonitorUri)->value_name("<uri>"),
+ "URI specifying monitor output (e.g. file:tmp/monitor.txt, empty = no monitor)")
("num-ts,n", po::value<int>(&fNumTimeslices)->default_value(-1)->value_name("<num>"),
"Stop after <num> timeslices (-1 = all)")
("skip-ts,s", po::value<int>(&fSkipTimeslices)->default_value(0)->value_name("<num>"),
diff --git a/algo/base/Options.h b/algo/base/Options.h
index 25c3f1b99a..61b480a4d2 100644
--- a/algo/base/Options.h
+++ b/algo/base/Options.h
@@ -22,6 +22,7 @@ namespace cbm::algo
const std::string& InputLocator() const { return fInputLocator; }
severity_level LogLevel() const { return fLogLevel; }
const std::string& Device() const { return fDevice; }
+ const std::string& MonitorUri() const { return fMonitorUri; }
bool CollectKernelTimes() const { return fCollectKernelTimes; }
int NumTimeslices() const { return fNumTimeslices; }
int SkipTimeslices() const { return fSkipTimeslices; }
@@ -31,6 +32,7 @@ namespace cbm::algo
std::string fInputLocator;
severity_level fLogLevel;
std::string fDevice;
+ std::string fMonitorUri;
bool fCollectKernelTimes = false;
int fNumTimeslices = -1;
int fSkipTimeslices = 0;
diff --git a/algo/base/SubChain.h b/algo/base/SubChain.h
index b787c1b64c..b206247190 100644
--- a/algo/base/SubChain.h
+++ b/algo/base/SubChain.h
@@ -18,6 +18,14 @@ namespace cbm::algo
const Options& Opts() const { return gsl::make_not_null(fContext)->opts; }
const RecoParams& Params() const { return gsl::make_not_null(fContext)->recoParams; }
+ bool HasMonitor() const { return gsl::make_not_null(fContext)->monitor.has_value(); }
+
+ Monitor& GetMonitor() const
+ {
+ // Need Get-prefix to avoid conflict with Monitor-class name
+ return gsl::make_not_null(fContext)->monitor.value();
+ }
+
private:
ChainContext* fContext = nullptr;
};
diff --git a/algo/base/util/TsUtils.h b/algo/base/util/TsUtils.h
new file mode 100644
index 0000000000..fa45dc060b
--- /dev/null
+++ b/algo/base/util/TsUtils.h
@@ -0,0 +1,19 @@
+/* Copyright (C) 2023 FIAS Frankfurt Institute for Advanced Studies, Frankfurt / Main
+ SPDX-License-Identifier: GPL-3.0-only
+ Authors: Felix Weiglhofer [committer] */
+
+#include <Timeslice.hpp>
+
+namespace cbm::algo::ts_utils
+{
+
+ inline size_t SizeBytes(const fles::Timeslice& ts)
+ {
+ size_t size = 0;
+ for (size_t i = 0; i < ts.num_components(); i++) {
+ size += ts.size_component(i);
+ }
+ return size;
+ }
+
+} // namespace cbm::algo::ts_utils
diff --git a/algo/global/Reco.cxx b/algo/global/Reco.cxx
index cea651468a..9984134c92 100644
--- a/algo/global/Reco.cxx
+++ b/algo/global/Reco.cxx
@@ -3,11 +3,14 @@
Authors: Felix Weiglhofer [committer] */
#include "Reco.h"
+#include <System.hpp>
+
#include <xpu/host.h>
#include "config/Yaml.h"
#include "log.hpp"
#include "util/TimingsFormat.h"
+#include "util/TsUtils.h"
using namespace cbm::algo;
@@ -26,6 +29,11 @@ void Reco::Init(const Options& opts)
xpu::device_prop props {xpu::device::active()};
L_(info) << "Running CBM Reco on Device " << props.name();
+ if (!opts.MonitorUri().empty()) {
+ fContext.monitor.emplace(opts.MonitorUri());
+ L_(info) << "Monitoring enabled, sending to " << opts.MonitorUri();
+ }
+
// Reco Params
fs::path recoParamsPath = opts.ParamsDir() / "RecoParams.yaml";
YAML::Node yaml = YAML::LoadFile(recoParamsPath.string());
@@ -76,6 +84,9 @@ void Reco::Run(const fles::Timeslice& ts)
xpu::timings ts_times = xpu::pop_timer();
+ GetMonitor().QueueMetric("cbm_reco", {{"hostname", fles::system::current_hostname()}},
+ {{"tsProcessed", 1}, {"bytesProcessed", ts_utils::SizeBytes(ts)}});
+
PrintTimings(ts_times);
}
--
GitLab