From 6db1c9c1d8b5eeba9fd522386fa41ba70c31b57f Mon Sep 17 00:00:00 2001
From: Thomas Applencourt <tapplencourt@anl.gov>
Date: Mon, 24 Jun 2024 16:17:46 -0500
Subject: [PATCH] more control of FS (#243)

Co-authored-by: Thomas Applencourt <applenco@aurora-uan-0010.hostmgmt.cm.aurora.alcf.anl.gov>
---
 .github/workflows/presubmit.yml |  1 +
 xprof/sync_daemon_fs            |  2 +-
 xprof/xprof.rb.in               | 30 +++++++++++++++++++++++-------
 3 files changed, 25 insertions(+), 8 deletions(-)

diff --git a/.github/workflows/presubmit.yml b/.github/workflows/presubmit.yml
index c0d8e3e7..ac58aa04 100644
--- a/.github/workflows/presubmit.yml
+++ b/.github/workflows/presubmit.yml
@@ -145,6 +145,7 @@ jobs:
         bats_file: ['mpi_sync_daemon', 'mpi_iprof']
         include:
           - bats_file: 'iprof'
+            thapi_sync_daemon: 'fs' # Need to force, if not `THAPI_SYNC_DAEMON` will be set to empty list
     env:
       THAPI_TEST_BIN: clinfo
       THAPI_BIN_DIR: ./build/ici/bin
diff --git a/xprof/sync_daemon_fs b/xprof/sync_daemon_fs
index 2d3fe45e..affe14d3 100755
--- a/xprof/sync_daemon_fs
+++ b/xprof/sync_daemon_fs
@@ -11,7 +11,7 @@ require 'etc'
 
 FOLDER_JOBID = File.join('.thapi_lock', mpi_job_id)
 SHARED_LOCAL_FILESYSTEM = File.join('/', 'dev', 'shm', Etc.getlogin, FOLDER_JOBID)
-SHARED_GLOBAL_FILESYSTEM = File.join(env_fetch_first('HOME'), FOLDER_JOBID)
+SHARED_GLOBAL_FILESYSTEM = File.join(env_fetch_first('THAPI_SHARED_PATH', 'HOME'), FOLDER_JOBID)
 
 # Use a log distribution seem to be a good tradeoff
 # between being nice to the FileSystem (not to many call)
diff --git a/xprof/xprof.rb.in b/xprof/xprof.rb.in
index 31271ba3..0be9c2f9 100755
--- a/xprof/xprof.rb.in
+++ b/xprof/xprof.rb.in
@@ -30,9 +30,9 @@ require 'date'
 module LoggerRefinement
   refine(Logger) do
     def info_block(message, &block)
-      info(message + ': entry') if message
+      info("#{message}: entry") if message
       block.call
-      info(message + ': exit') if message
+      info("#{message}: exit") if message
     end
   end
 end
@@ -259,14 +259,28 @@ class Sync_daemon
   end
 
   def initialize
-    daemon_type = env_fetch_first('THAPI_SYNC_DAEMON', default: 'mpi')
-    lazy_exec("Initialize Sync_daemon #{daemon_type}") do
-      daemon = if File.exist?("#{__dir__}/sync_daemon_mpi") && daemon_type == 'mpi'
+    daemon_type = env_fetch_first('THAPI_SYNC_DAEMON')
+    daemon = case daemon_type
+             when nil
+               # Default is MPI
+               if File.exist?("#{__dir__}/sync_daemon_mpi")
                  "#{__dir__}/sync_daemon_mpi"
                else
+                 LOGGER.warn("No #{__dir__}/sync_daemon_mpi binary. Fall back to #{__dir__}/syc_daemon_f'")
                  "#{__dir__}/sync_daemon_fs"
                end
-      LOGGER.debug { "spawn(#{daemon} #{Process.pid})" }
+             when 'mpi'
+               raise("No #{__dir__}/sync_daemon_mpi binary") unless File.exist?("#{__dir__}/sync_daemon_mpi")
+
+               "#{__dir__}/sync_daemon_mpi"
+             when 'fs'
+               "#{__dir__}/sync_daemon_fs"
+             else
+               raise("Error: THAPI_SYNC_DAEMON value (#{daemon_type}) if not supported. Allowed: [mpi,fs] ")
+             end
+
+    LOGGER.debug { "spawn(#{daemon} #{Process.pid})" }
+    lazy_exec("Initialize Sync_daemon #{daemon_type}") do
       @pid = spawn("#{daemon} #{Process.pid}")
     end
   end
@@ -293,8 +307,10 @@ class Sync_daemon
   # we always call clean-up the daemon
   def self.open
     yield f = new
+  rescue StandardError
+    raise
   ensure
-    f.finalize
+    f.finalize if f
   end
 end