From 26890325c35a5a57b07f25be99aaddf211a75255 Mon Sep 17 00:00:00 2001
From: Chris Jones <jonesc@hep.phy.cam.ac.uk>
Date: Tue, 17 Jan 2023 13:12:44 +0000
Subject: [PATCH] ProcStats: Attempt to reconnection to process proc stat when
information appeats corrupted
---
GaudiAud/src/ProcStats.cpp | 29 +++++++++++++++++------------
GaudiAud/src/ProcStats.h | 13 ++++++++-----
2 files changed, 25 insertions(+), 17 deletions(-)
diff --git a/GaudiAud/src/ProcStats.cpp b/GaudiAud/src/ProcStats.cpp
index a96beee71e..386115a512 100644
--- a/GaudiAud/src/ProcStats.cpp
+++ b/GaudiAud/src/ProcStats.cpp
@@ -30,10 +30,6 @@
# endif // __linux__
# include <cstdio>
-using std::cerr;
-using std::cout;
-using std::endl;
-
/* Format of the Linux proc/stat (man 5 proc, kernel 2.6.35):
pid %d The process ID.
@@ -245,35 +241,39 @@ ProcStats* ProcStats::instance() {
return &inst;
}
-ProcStats::ProcStats() {
+void ProcStats::open_ufd() {
+ m_valid = false;
#if defined( __linux__ ) or defined( __APPLE__ )
+ m_ufd.close();
m_pg_size = sysconf( _SC_PAGESIZE ); // getpagesize();
const auto fname = "/proc/" + std::to_string( getpid() ) + "/stat";
m_ufd.open( fname.c_str(), O_RDONLY );
if ( !m_ufd ) {
- cerr << "Failed to open " << fname << endl;
- return;
+ std::cerr << "ProcStats : Failed to open " << fname << std::endl;
+ } else {
+ m_valid = true;
}
#endif // __linux__ or __APPLE__
- m_valid = true;
}
+ProcStats::ProcStats() { open_ufd(); }
+
bool ProcStats::fetch( procInfo& f ) {
if ( !m_valid ) { return false; }
- std::scoped_lock lock{ m_mutex };
+ std::scoped_lock lock{m_mutex};
#if defined( __linux__ ) or defined( __APPLE__ )
- double pr_size{ 0 }, pr_rssize{ 0 };
+ double pr_size{0}, pr_rssize{0};
linux_proc pinfo;
- int cnt{ 0 };
+ int cnt{0};
char buf[500];
m_ufd.lseek( 0, SEEK_SET );
if ( ( cnt = m_ufd.read( buf, sizeof( buf ) ) ) < 0 ) {
- cout << "LINUX Read of Proc file failed:" << endl;
+ std::cerr << "ProcStats : LINUX Read of Proc file failed:" << std::endl;
return false;
}
@@ -310,5 +310,10 @@ bool ProcStats::fetch( procInfo& f ) {
m_curr.rss = f.rss;
m_curr.vsize = f.vsize;
+ if ( 0 == m_curr.vsize ) {
+ std::cerr << "ProcStats : 0==vsize -> Will try reopening process proc stats" << std::endl;
+ open_ufd();
+ }
+
return rc;
}
diff --git a/GaudiAud/src/ProcStats.h b/GaudiAud/src/ProcStats.h
index 0452127134..27c22ea08f 100644
--- a/GaudiAud/src/ProcStats.h
+++ b/GaudiAud/src/ProcStats.h
@@ -46,8 +46,8 @@ struct procInfo {
}
// see proc(4) man pages for units and a description
- double vsize{ 0 }; // in MB (used to be in pages?)
- double rss{ 0 }; // in MB (used to be in pages?)
+ double vsize{0}; // in MB (used to be in pages?)
+ double rss{0}; // in MB (used to be in pages?)
};
class ProcStats {
@@ -62,7 +62,7 @@ private:
class unique_fd {
private:
- int m_fd{ -1 };
+ int m_fd{-1};
unique_fd( const unique_fd& ) = delete;
unique_fd& operator=( const unique_fd& ) = delete;
@@ -107,10 +107,13 @@ private:
#undef unique_fd_forward
};
+private:
+ void open_ufd();
+
private:
unique_fd m_ufd;
- double m_pg_size{ 0 };
+ double m_pg_size{0};
procInfo m_curr;
- bool m_valid{ false };
+ bool m_valid{false};
std::mutex m_mutex;
};
--
GitLab