Scribe: The record-replay mechanism
Scribe is a low-overhead multi-threaded application record-replay mechanism.
Scribe introduces new lightweight operating system mechanisms, rendezvous and sync points, to efficiently record nondeterministic interactions such as related system calls, signals, and shared memory accesses. Rendezvous points make a partial ordering of execution based on system call dependencies sufficient for replay, avoiding the recording overhead of maintaining an exact execution ordering. Sync points convert asynchronous interactions that can occur at arbitrary times into synchronous events that are much easier to record and replay.
For more details read this blog post. It has a screencast showing some of Scribe features.
The Scribe project is divided in four different ones:
- GCC and its friends
- Python (version 2.6)
Install the kernel
git clone git://github.com/nviennot/linux-2.6-scribe.git cd linux-2.6-scribe make menuconfig make make install
Install the C library
git clone git://github.com/nviennot/libscribe.git cd libscribe cd build cmake .. make install
Install the python library and userspace tools
git clone git://github.com/nviennot/py-scribe.git cd py-scribe ./setup install
(Optional) Install the test suite
git clone git://github.com/nviennot/tests-scribe.git
py-scribe provides three scripts: record, replay, profiler.
1. Record an application
The record command line tool allows the user to record an execution.
The verbosity level of the recorded log file can be provided. It allows the user to record only the bare minimum to guarantee a deterministic replay (highest performance), or to record the execution with debugging information so the log file can be easily interpreted by getting a execution trace similar to strace.
By sending a
SIGUSR1 signal to the recording tool, Scribe detaches itself
from the application while it continues running.
SIGUSR2 signal bookmarks an execution point in time. The user can then
replay the application up to that point and the application state will be
guaranteed to be exactly the same as during the recording.
# record date Mon Aug 8 04:18:33 EDT 2011 # ls -lh date.log -rw-r--r-- 1 root root 4.2K Aug 8 04:18 date.log
2. Replay an execution from a log file
The replay tool allows the user to replay a previously recorded execution.
The user can provide the backtrace size in case the replay fails and diverge (for instance, the system got out of memory and the replay cannot continue).
SIGUSR1 signal can be sent to detach Scribe at any point in time and let the
application continue a normal execution.
A bookmark id can be given as well to let the application go live at a specific point in time.
# replay date.log Mon Aug 8 04:18:33 EDT 2011
3. Look at the recorded log file in a human readable format
The profiler tool allows the user to display the recorded log file in a human readable format.
# profiler date.log | grep Mon -B3  write() = 29  resource lock, type = files_struct, serial = 31  resource lock, type = file, serial = 1, desc = /dev/pts/0  data: size = 29, "Mon Aug 8 04:18:33 EDT 2011\n"
The provided command line tools use the Scribe Python library internally. The user can use the libraries to achieve a lot more by building its own logic around the Scribe API.
- For the kernel implementation details, read the scribe kernel documentation (OUTDATED).