================================================== Ocean Observatories Initiative Cyberinfrastructure LCAarch - OOI Release 1 LCA architecture prototype ================================================== April 2010 - August 2010 (C) UCSD Regents This project provides a service framework with auxilliary functions for running architecturally complete versions of all the services of the OOI release 1 system with their full architectural dependencies in Python. Functionally it provides a data distribution network based on pub-sub messaging as well as instrument integration, with underlying infrastructure services (security, persistence) and provisioning. The "core" part of LCAarch provides base classes and a framework for spawnable, message communicating processes, for services with defined names, for bootstrapping the system, for managing logging and configuration etc. This is an early implementation of the OOI Python Capability Container. For more information, please see: http://www.oceanobservatories.org/spaces/display/CIDev/LCAARCH+Development+Project Get LCAarch with :: git clone git@amoeba.ucsd.edu:lcaarch.git cd lcaarch Dependencies ============ LCAarch is compatible with Python 2.5 and higher, but not Python 3.x This project has several dependencies on libraries and packages. Most of these dependencies are resolved automatically using the provided setup script. Step 1: Virtual env Create a Python virtual environment using virtualenv. This ensures that all libraries and dependencies are installed separately from the Python system libraries. :: mkvirtualenv lcaarch workon lcaarch Step 2: Core libraries (you can skip this step) Install some core libraries first. Sometimes the automatic installer produces errors, if these libraries are not present beforehand. :: easy_install twisted numpy Step 3: Run the setup script :: python setup.py install This should download and install all the dependencies and will run for a while. Check the trace output that there are no substantial errors. You are now ready to run. Current dependencies include: twisted, numpy, txamqp, msgpack-python, httplib2, pycassa, simplejson, pydap, pydap.handlers.netcdf, pydap.responses.netcdf, pydap.handlers.nca, gviz_api.py NOTE: As the project evolves and new code is added, dependencies might change. Run the setup script once in a while and when you get errors Usage ===== (all subsequent steps assume you are in the lcaarch/ root dir) Start empty CC ("Magnet" Python Capability Container) shell with: :: twistd -n magnet -h amoeba.ucsd.edu twistd -n magnet # to run with localhost (to end a magnet container shell, press Ctrl-D Ctrl-C) Start system by executing within the CC shell: ><> from ion.core import bootstrap bootstrap.start() Alternatively (better) from UNIX shell executing a script: :: twistd -n magnet -h amoeba.ucsd.edu res/scripts/bootstrap.py twistd -n magnet -h amoeba.ucsd.edu res/scripts/newcc.py twistd -n magnet -h amoeba.ucsd.edu -a nproducers=25 res/scripts/pubsub.py Testing ======= Run trial test cases (recursively) :: trial ion trial ion.core trial ion.services.coi.test.test_resource_registry A good learning example is the HelloService :: trial ion.play.test.test_hello Or in the CC shell: ><> from ion.play import hello_service spawn(hello_service) send(1, {'op':'hello','content':'Hello you there!'}) from ion.play.hello_service import HelloServiceClient hc = HelloServiceClient() hc.hello() Build and Packaging using Ant ============================= LCAarch provides ANT support (see http://ant.apache.org/). To check that ant is installed properly, run :: ant To clean your working directories, run :: ant clean To compile all code to see if there are Python compile errors anywhere: :: ant compile --------------------------------------------------------------------------- Change log: =========== 2010-08-14: - BaseProcess: added backend receiver, used for sending out any messages from self.send and self.rpc_send. This keeps the message queue for the process frontend separate from the process backend, e.g. for RPC during a message processing. - Changed BaseProcess logging to make message send and receive easier to spot. 2010-08-06: - BaseProcess.spawn() now calls init() automatically. No need to call init() on a process anymore manually. For testing only. Note: The correct way to spawn a process is through a parent process with spawn_child() - Modified and fixed the BaseProcess states, when receiving messages - MAJOR update to BaseProcess message dispatching and subsequent error handling. On error, reply_err messages are sent back, if reply-to header set. 2010-08-03: - Added ant build.xml file to LCAarch root dir. Start with ant. Supports ant clean, which removes all *.pyc from ion path. 2010-07-23: - Refactored the Registry Services to inherit from a common base class. This will allow easier implementation of the many registries in the OOICI. The Resource Registry and Service Registry now have basic registration of resource descriptions and services. - The Ion message is now encoded usig the MsgPack library to allow for binary transport of message content. The JSON library was mangleing string content. This is a temporary fix for a wider problem with the encoding structure which needs to be addressed in construction. 2010-06-07: - Redefined logging import to set as module level logger. Each module has now a different logger that can be configured separately. Don't configure manually though. - Added possibility to modify logging levels by module/package, with support for package hierarchy. See res/logging/loglevels.cfg for standard entries. Do not modify. Add file res/logging/loglevelslocal.cfg for local override. Default logging level for all ion code: WARNING 2010-06-02: - BaseProcess self members renamed to conform to PEP8 - Added process shutdown to BaseProcess - Added container UNIX shell argument -a processes=<path to filename> used by newcc.py script, with a list of processes to startup in standard format 2010-05-25: - Made Cassandra backend parameterizable with keyspace/colfamily and added SuperColumn support. - Modified the IStore interface to support a create_store factory method. This method can yield and return a deferred. Modified and fixed IStore impls. Changed delete to remove to be more compliant with standard collections. 2010-05-22: - Added timeout to BaseProcess.rpc_send. Use with kwarg timeout=<secs> - CC-Agent detects missing known containers and removes them from the list - Enhanced CC-Agent operations and CC shell helpers - Added sequence numbers for messages - Added glue functions BaseProcess.reply_ok and reply_err and changes some RPC style service operations. 2010-05-20: - The system now looks for a local config file ionlocal.config and if exists, overrides entries in ion.config. - Test cases use the config file to determine the broker host to use. If local config override exists, a different broker (e.g. localhost) can be given. - Added BaseProcessClient and changed BaseServiceClient and all clients and all test cases (again). - Added container shell helpers under 'cc', such as cc.spawn('hello') 2010-05-16: - Removed support for BaseProcess.send_message and reply_message. Always use send, reply and rpc_send now. - Any BaseProcess instance can now spawn_child() other processes. - Removed RpcClient class, because every process can do rpc_send() - Service processes now also listen to their service name's queue. The service name is determined from the service declaration. Two processes will listen to the same queue and take messages round robin from the queue. - Startup arguments evaluated, for instance to start with system name set: twistd -n magnet -a sysname=mysys twistd -n magnet -a "{'sysname':'mysys'}" - Added capability container agent process. Start with: twistd -n magnet res/scripts/newcc.py Agents announce themselves to others in the same system and can spawn procs. - Name scope 'local' for messaging names means now really local to one container. Use scope 'system' for names unique for each bootstrapped system. Do not use global names, because they will clash. - Less verbose trace output for process init messages and changes to other trace output as well. - Changed BaseServiceClient and tests. Initializer arguments different. - Using master branch of magnet now 2010-05-10: - Based on entries in config files, service process modules are sought and loaded in order to collect the service process declarations. This enables a registration of services with versions and dependencies on startup 2010-05-06: - Refactored the BaseProcess class. - You can now do RPC directly from the process, via self.rpc_send without the need for an RpcClient. Works even with other messages coming in at the same time (using the conv-id) - Added aliases: self.send, self.reply (use these now) - Process instance can be spawned via self.spawn() - Provided an easier to use BaseServiceClient, which uses a default service name lookup. Accepts BaseProcess instance as argument to use for sending/ receiving service calls.