Debugger Manual

1 Introduction

The primary goal of the parallel debugger is to provide an integrated debugging environment which allows the programmer to examine the changing state of the parallel program during the course of its execution.

The Charm++ debugging system has a number of useful features for Charm++ programmers. The system includes a Java GUI client program which runs on the programmer's desktop, and a Charm++ parallel program which acts as a server. The client and server need not be on the same machine, and communicate over the network using a secure protocol described in

The system provides the following features:

The debugging client provides these features via extensive support built into the Charm++ runtime.

2 Building the Charm++ Debug Tool

To get the Charm Debug tool, check out the source code from the repository. This will create a directory named ccs_tools. Move to this directory and build Charm Debug.

 git clone
 cd ccs_tools

This will create the executable bin/charmdebug, which should work.

You can also download the binaries from the Charm++ downloads website and use it directly without building. (NOTE: Binaries may not be properly working in some platforms, so building from the source code is recommended.)

3 Preparing the Charm++ Application for Debugging

Build Charm++ using -enable-charmdebug option. For example:

  ./build charm++ netlrts-darwin-x86_64 --enable-charmdebug

No instrumentation is required to use the Charm++ debugger. Being CCS based, you can use it to set and step through entry point breakpoints and examine Charm++ structures on any Charm++ application.

Nevertheless, for some features to be present some additional options might be required at either compile or link time:

3.1 Record Replay

The Record Replay feature is independent of the charmdebug application. It is a mechanism used to detect bugs that happen only once in a while depending on the order in which messages are processed. The program in consideration is first run in a record mode which produces a trace. When the program is run in replay mode it uses a previous trace gotten from a record run to ensure that messages are processed in the same order as the recorded run. The idea is to make use of a message-sequence number and a theorem says that the serial numbers will be the same if the messages are processed in the same order. [#!rashmithesis!#]

Record Replay tracing is automatically enabled for Charm++ programs and requires nothing special to be done during compilation (linking with the option ``-tracemode recordreplay'' used to be necessary). At run time, the ``+record'' option is used, which records messages in order in a file for each processor. The same execution order can be replayed using the ``+replay'' runtime option, which can be used at the same time as the other debugging tools in Charm++ .

Note! If your Charm++ is built with CMK_OPTIMIZE on, all tracing will be disabled. So, use an unoptimized Charm++ to do your debugging.

4 Running the Debugger

4.1 Charmdebug command line parameters

Number of PEs

Number of PEs

hostname of CCS server for application

the username to use to connect to the hostname selected

portnumber of CCS server for application

force the communication between client and server (in particular the one for CCS) to be tunnelled through ssh. This allow the bypass of firewalls.

X Display

4.2 Basic usage

To run an application locally via the debugger on 4 pes with command line options for your pgm (e.g. opt1 opt2 ):

  charmdebug pgm +p4 4 opt1 opt2

If the application should be run in a remote cluster behind a firewall, the previous command line will become:

  charmdebug -host -user myname -sshtunnel pgm +p4 4 opt1 opt2

Charmdebug can also be executed without any parameters. The user can then choose the application to launch and its command line parameters from within the File menu as shown in Figure 1.

Figure 1: Using the menu to set parameters for the Charm++ program
Image menu

Note: charmdebug command line launching only works on netlrts-* and verbs-* builds of Charm++

To replay a previously recorded session:

  charmdebug pgm +p4 opt1 opt2  +replay

4.3 Charm Debugging Related Options

When using the charm debugger to launch your application, it will automatically set these to defaults appropriate for most situations.

Triggers application freeze at startup for debugger.

Triggers charmrun to provide some information about the executable, as well as provide an interface to gdb for querying.

Which debuggers to use.

Run each node under gdb in an xterm window, prompting the user to begin execution.

Run each node under gdb in an xterm window immediately (i.e. without prompting the user to begin execution).

Note: If you're using the charm debugger it will probably be best to control the sequential (i.e. gdb) debuggers from within its GUI interface.

X Display for xterm

Port to listen for CCS requests

Enable client-server (CCS) mode

Use the recordreplay tracemode to record the exact event/message sequence for later use.

Force the use of recorded log of events/messages to exactly reproduce a previous run.

The preceding pair of commands +record +replay are used to produce the ``instant replay'' feature. This feature is valuable for catching errors which only occur sporadically. Such bugs which arise from the nondeterminacy of parallel execution can be fiendishly difficult to replicate in a debugging environment. Typical usage is to keep running the application with +record until the bug occurs. Then run the application under the debugger with the +replay option.

4.4 Charmdebug limitations

4.4.1 Clusters

Charmdebug is currently limited to applications started directly by the debugger due to implementation peculiarities. It will be extended to support connection to remote running applications in the near future.

Due to the current implementation, the debugging tool is limited to netlrts-* and verbs-* versions. Other builds of Charm++ might have unexpected behavior. In the near future this will be extended at least to the mpi-* versions.

4.4.2 Record Replay

The record replay feature does not work well with spontaneous events. Load balancing is the most common form of spontaneous event in that it occurs periodically with no other causal event. As per

Figure 2: Parallel debugger when a break point is reached
Image snapshot3

As per Rashmi's thesis: There are some unique issues for replay in the context of Charm because it provides high-level support for dynamic load balancing, quiescence detection and information sharing. Many of the load balancing strategies in Charm have a spontaneous component. The strategy periodically checks the sizes of the queues on the local processor. A replay load balancing strategy implements the known load redistribution. The behavior of the old balancing strategy is therefore not replayed only its effect is. Since minimal tracing is used by the replay mechanism the amount of perturbation due to tracing is reduced. The replay mechanism is proposed as a debugging support to replay asynchronous message arrival orders.

Moreover, if your application crashes without a clean shutdown, the log may be lost with the application.

4.5 Using the Debugger

Once the debugger's GUI loads, the programmer triggers the program execution by clicking the Start button. When starting by command line, the application is automatically started. The program starts off displaying the user and system entry points as a list of check boxes, freezing at the onset. The user could choose to set breakpoints by clicking on the corresponding entry points and kick off execution by clicking the Continue Button. Figure 2 shows a snapshot of the debugger when a breakpoint is reached. The program freezes when a breakpoint is reached.

Clicking the Freeze button during the execution of the program freezes execution, while Continue button resumes execution. Quit button can be used to abort execution at any point of time. Entities (for instance, array elements) and their contents on any processor can be viewed at any point in time during execution as illustrated in Figure 3.

Figure 3: Freezing program execution and viewing the contents of an array element using the Parallel Debugger
Image arrayelement

Specific individual processes of the Charm++ program can be attached to instances of gdb as shown in Figure 4. The programmer chooses which PEs to connect gdb processes to via the checkboxes on the right side. Note! While the program is suspended in gdb for step debugging, the high-level features such as object inspection will not work.

Figure 4: Parallel debugger showing instances of gdb open for the selected processor elements
Image snapshot4-crop

Charm++ objects can be examined via the View Entities on PE : Display selector. It allows the user to choose from Charm Objects, Array Elements, Messages in Queue, Readonly Variables, Readonly Messages, Entry Points, Chare Types, Message Types and Mainchares. The right sideselector sets the PE upon which the request for display will be made. The user may then click on the Entity to see the details.

4.5.1 Memory View

The menu option Action $\rightarrow$ Memory allows the user to display the entire memory layout of a specific processor. An example is shown in figure 5. This layout is colored and the colors have the following meaning:

Figure 5: Main memory view
Image memoryView

memory allocated by the Charm++ Runtime System;

memory allocated directly by the user in its code;

memory used by messages;

memory allocated to a chare element;

memory not allocated;

a big jump in memory addresses due to the memory pooling system, it represent a large portion of virtual space not used between two different zones of used virtual space address;

the currently selected memory slot;

Currently it is not possible to change this color association. The bottom part of the view shows the stack trace at the moment when the highlighted (yellow) memory slot was allocated. By left clicking on a particular slot, this slot is fixed in highlight mode. This allows a more accurate inspection of its stack trace when this is large and does not fit the window.

Info $\rightarrow$Show Statistics will display a small information box like the one in Figure 6.

Figure 6: Information box display memory statistics
Image memoryStatistics

A useful tool of this view is the memory leak search. This is located in the menu Action $\rightarrow$ Search Leaks. The processor under inspection runs a reachability test on every memory slot allocated to find if there is a pointer to it. If there is none, the slot is partially colored in green, to indicate its status of leak. The user can the inspect further these slots. Figure 7 shows some leaks being detected.

Figure 7: Memory view after running the Search Leaks tool
Image memoryLeaking

If the memory window is kept open while the application is unfrozen and makes progress, the loaded image will become obsolete. To cope with this, the ``Update'' button will refresh the view to the current allocation status. All the leaks that had been already found as such, will still be partially colored in green, while the newly allocated slots will not, even if leaking. To update the leak status, re-run the Search Leaks tool.

Finally, when a specific slot is highlighted, the menu Action $\rightarrow$ Inspect opens a new window displaying the content of the memory in that slot, as interpreted by the debugger (see next subsection for more details on this).

4.5.2 Inspector framework

Without any code rewriting of the application, CharmDebug is capable of loading a raw area of memory and parse it with a given type name. The result (as shown in Fig. 8), is a browseable tree. The initial type of a memory area is given by its virtual table pointer (Charm++ objects are virtual and therefore loadbable). In the case of memory slots not containing classes with virtual methods, no display will be possible.

Figure 8: Raw memory parsed and displayed as a tree
Image memoryInspector

When the view is open and is displaying a type, by right clicking on a leaf containing a pointer to another memory location, a popup menu will allow the user to ask for its dereference (shown in Fig. 8). In this case, CharmDebug will load this raw data as well and parse it with the given type name of the pointer. This dereference will be inlined and the leaf will become an internal node of the browse tree.

5 Debugger Implementation Details

The following classes in the PUP framework were used in implementing debugging support in charm.

The code below shows a simple class declaration that includes a pup method.

  class foo {
    bool isBar;
    int x;
    char y;
    unsigned long z;
    float q[3];
    void pup(PUP::er &p) {

5.1 Converse Client-Server Interface

The Converse Client-Server (CCS) module enables Converse [#!InterOpIPPS96!#] programs to act as parallel servers, responding to requests from non-Converse programs. The CCS module is split into two parts - client and server. The server side is used by a Converse program while the client side is used by arbitrary non-Converse programs. A CCS client accesses a running Converse program by talking to a server-host which receives the CCS requests and relays them to the appropriate processor. The server-host is charmrun [#!charmman!#] for netlrts- versions and is the first processor for all other versions.

In the case of the netlrts- version of Charm++ , a Converse program is started as a server by running the Charm++ program using the additional runtime option ``++server''. This opens the CCS server on any TCP port number. The TCP port number can be specified using the command-line option ``server-port''. A CCS client connects to a CCS server, asks a server PE to execute a pre-registered handler and receives the response data. The function CcsConnect takes a pointer to a CcsServer as an argument and connects to the given CCS server. The functions CcsNumNodes, CcsNumPes, CcsNodeSize implemented as part of the client interface in Charm++ returns information about the parallel machine. The function CcsSendRequest takes a handler ID and the destination processor number as arguments and asks the server to execute the particular handler on the specified processor. CcsRecvResponse receives a response to the previous request in-place. A timeout is also specified which gives the number of seconds to wait till the function returns a 0, otherwise the number of bytes received is returned.

Once a request arrives on a CCS server socket, the CCS server runtime looks up the appropriate registered handler and calls it. If no handler is found the runtime prints a diagnostic and ignores the message. If the CCS module is disabled in the core, all CCS routines become macros returning 0. The function CcsRegisterHandler is used to register handlers in the CCS server. A handler ID string and a function pointer are passed as parameters. A table of strings corresponding to appropriate function pointers is created. Various built-in functions are provided which can be called from within a CCS handler. The debugger behaves as a CCS client invoking appropriate handlers which makes use of some of these functions. Some of the built-in functions are as follows.

The CCS runtime system provides several built-in CCS handlers, which are available to any Converse program. All Charm++ programs are essentially Converse programs. ccs_getinfo takes an empty message and responds with information about the parallel job. Similarly the handler ccs_killport allows a client to be notified when a parallel run exits.