Upstream version 3.3.0~rc5

Stéphane Martin authored 10 years ago and JEANVOINE Emmanuel committed 10 years ago

Before this patch the cache had useless functions and the object FileCache
could be re-index in the Cache data structure.  The size was not recomputed and
there was race condition on it.

This patch uses the monitor design pattern to simplify the design.  The cache
has a lock to add or remove elements. Every FileCache has a lock to update, acquire or
release a token, fetch a file etc.

This patch updates test/test_cache.rb to test this module.

Change-Id: I9a660ce179d090d4c449c40ca5a7b1a7b82707eb

Change-Id: I4cebfaf1da713f2b5ada6e95df5964027920fa6e

Since ruby 2.0, it is forbiden to take a lock or synchronize inside a trap context.
To resolve this issue, this patch adds thread inside trap and uses the thread context.

Try this code with thread arguement or not in different ruby version:

$mumu = Mutex.new
def fault()
  begin
    $mumu.synchronize do
      puts "int"
    end
  rescue Exception => ex
     puts ex
     puts ex.backtrace
  end
end
def thread_fault()
  t=Thread.new do
    fault()
  end
  t.join
end
if ARGV[0] == 'thread'
  Signal::trap("INT") do
    thread_fault()
  end
else
  Signal::trap("INT") do
    fault()
  end
end
Process.kill("INT",$$)

Change-Id: If5130a2369e2d2a15a7081d421ad52d5c88fe130

Stéphane Martin authored 10 years ago and Lucas Nussbaum committed 10 years ago

Change-Id: I06be7d992d9048620619d3dac9085a39e5ac1e2c

Merge "[Kadeploy3d] Bugfix: Use thread context instead of trap context for kill kadeploy for INT and TERM with ruby >=2.0."

Stéphane Martin authored 10 years ago and Lucas Nussbaum committed 10 years ago

Change-Id: I8723dc9f7ac6fb68b8155945ec0ec086a5b974f9

Stéphane Martin authored 10 years ago and Lucas Nussbaum committed 10 years ago

When a USR2 signal was received, Kadeploy explores all thread and display a backtrace.
A backtrace of a thread could be nil. Before this patch, we try to call join even if this backtrace is nil.

This patch resolve this issue.

See bug https://intranet.grid5000.fr/bugzilla/show_bug.cgi?id=5650 for
details.

Change-Id: Iac7831450a983aaa3e52e71266285e87f44ce057

[Kadeploy3d] Bugfix: Use thread context instead of trap context for kill kadeploy for INT and TERM with ruby >=2.0.

Since ruby 2.0, it is forbiden to take a lock or synchronize inside trap context.
To resolve this issue, this patch adds thread inside trap and uses the thread context.

Try this code with thread arguement or not in different ruby version:
$mumu = Mutex.new
def fault()
  begin
    $mumu.synchronize do
      puts "int"
    end
  rescue Exception => ex
     puts ex
     puts ex.backtrace
  end
end

def thread_fault()
  t=Thread.new do
    fault()
  end
  t.join
end

if ARGV[0] == 'thread'
  Signal::trap("INT") do
    thread_fault()
  end
else
  Signal::trap("INT") do
    fault()
  end
end

Process.kill("INT",$$)

Change-Id: I4959d8d45ba74bb6b1452dac7c985daf619b9b02

Since ruby 2.0, the signal trap has special context.
This patch add a Thread to resolve this issue.

Change-Id: I365a4ff686a38600dbdc40e3d0db23e0ad1c25de

Before this patch, the end of database connection must be made in two times:
disconnect and free. The issue is that kaworkflow calls free without
disconnect.

Now free function closes the SQL connection if a connection was established
and disconnect function is deleted. This resolves the kaworkflow issue.

Note for developers:

Kadeploy establishes one SQL connexion on each action (in fact, each connexion
to server, it can be avoided for the polling if workflow keeps the user right
only for the polling).  When a workflow is created, a second SQL connection is
established because the workflow has longer life than execution context.

My opinion is to made only one connection for everything. I was tested with the
modification of file server.rb: check_database establishes the SQL connection
and database_handler return the connection. It works fine with concurrency.
But EJ said that the MySQL module has weird behavior after one week of running.

Change-Id: I2b1610890aded0b538fade4a3fc386857b777127

With this patch, there is only one thread to manage the pipes and waitpid.
The wait of this function checks the status of this thread.
Killing a 'execute.wait()' function kills the process with 'execute.kill()' function.

This patch provides some test for execute class in test/test_execute.rb.

Change-Id: Id585094f462c86647b8b4ef3013e64b00badcd91

When the client is called with multi-site the result function will be called
many times.  But the result files are opened with w+ and it overwrite the file.

This patch use a+ to append the file.  The file will be reseted on argument
parsing inside the load_outputfile function.

Change-Id: I4aaf313013631a9b1aa6fe90f2369924fa5bbd4a

Upstream version 3.3.0~rc4

Change-Id: I3b2f1f70ff4a158df08e746a818a3bd2b384df0b

Upstream version 3.3.0~rc3

The module Ruby/Mysql has a part in C language. This part manages the socket
which is connected to the server.  This socket is not referenced by ruby. When
we fork this socket is dumped to the child and we can not find this socket with
ObjectSpace.

So this patch replace ObjectSpace by "Dir.foreach('/proc/self/fd')". This line
iterate with all opened file descriptor for the current application.

Change-Id: I214638750dbdb6e7f3e1e6cd2768d1fcf959c072

The kaworkflow uses different matching than environment.
This is an issue when the API is used. Because the API use directly
deploy without ask environment

This patch fixes the bug: https://intranet.grid5000.fr/bugzilla/show_bug.cgi?id=5631

Change-Id: Idb3c7caabbf8dc550438c4afc81b5c223d7c079b

The webrick server has pool of threads. Threrefore, when a servlet never
finishes, the used thread becomes busy.  When all threads are busy, the webrick
server can't respond to new request.  This timeout raise an exception after 10
min if the servlet has not finished. This exception shows where the thread was
locked.

Change-Id: Ic646ab22c33760e242238f3a3551eec74dd7e5ce

Before this patch the hook part of kaworkflow catch SignalException to detect
if the timeout was reached.  Now we need to catch Timeout::Error, because the
exception SignalException is no more raised since the patch
https://intranet.grid5000.fr/gerrit/#/c/2198/
(61156f11)

Change-Id: I3333bd46b63a5e3969f298c32d07498ed73ee524

With this patch the arity and auto-propagate parameters of kadeploy is given to kascade

Change-Id: Id6143c7812d08dcacca848a6e0bef98bf9cd1d7d

Change-Id: I2dfe59dc453fef46438ba5c0875436030f42f3d8