This is a client package that allows connection to a database that stores telescope state information for the Science Data Processor of the MeerKAT radio telescope. This database is colloquially known as telstate.
The telescope state is a key-value store. There are three types of keys:
- immutables (aka attributes)
Stores a single value that is not allowed to change once set.
- mutables (aka sensors)
Stores multiple timestamped values organised into an ordered set.
- indexed
Stores a dictionary of key-value pairs, each of which behaves like an immutable. This is useful to avoid the main key-space becoming too large. It also supports some patterns like incrementally storing values but fetching all values in a single operation. Furthermore, it allows more general keys than just strings.
The keys are strings and the values (and the sub-keys of indexed keys) are Python objects serialised via MessagePack, which has been extended to support tuples, complex numbers and NumPy arrays. Older versions of the database stored the values as pickles, and the package warns the user if that's the case. Keys can be retrieved from the telstate object using attribute syntax or dict syntax.
Databases can be accessed via one of two backends: a Redis client backend that allows shared access to an actual Redis server over the network (or a simulated server via fakeredis) and a simplified in-memory backend for stand-alone access. Both backends support loading and saving a Redis snapshot in the form of an RDB dump file.
It is possible to have multiple views on the same database (one per telstate instance). A view is defined as a list of prefixes acting as namespaces that group keys. When reading from the database, each prefix is prepended to the key in turn until a match is found. When writing to the database, the first prefix is prepended to the key. The first prefix therefore serves as the primary namespace while the rest are supplementary read-only namespaces.
Warning
WARNING: The standard warning about Python pickles applies. Never retrieve data from an untrusted telstate database with values encoded as pickles, or connect to such a database over an untrusted network. Pickle support is disabled by default, but can be enabled for trusted databases by setting the environment variable KATSDPTELSTATE_ALLOW_PICKLE=1.
The simplest way to test out katsdptelstate is to use the in-memory backend. If you want to run a real Redis server you will need to install Redis (version 4.0 or newer) on a suitable machine on the network. For example, do this:
- macOS:
brew install redis
- Ubuntu:
apt-get install redis-server
Then pip install katsdptelstate
and run a local redis-server
. If you also want to load RDB files, do pip install katsdptelstate[rdb]
.
import time
import katsdptelstate
# Connect to an actual Redis server via an endpoint or an URL
telstate = katsdptelstate.TelescopeState('localhost:6379')
telstate = katsdptelstate.TelescopeState('redis://localhost')
# Or use the in-memory backend (useful for testing)
telstate = katsdptelstate.TelescopeState()
# Load RDB file into Redis if katsdptelstate is installed with [rdb] option
telstate.load_from_file('dump.rdb')
# Attribute / dict style access returns the latest value
telstate.add('n_chans', 32768)
print(telstate.n_chans) # -> 32768
print(telstate['n_chans']) # -> 32768
# List all keys (attributes and sensors)
print(telstate.keys()) # -> ['n_chans']
# Sensors are timestamped underneath
st = time.time()
telstate.add('n_chans', 4096)
et = time.time()
telstate.add('n_chans', 16384)
# Time ranges can be used and are really fast
telstate.get_range('n_chans', st=st, et=et) # -> [(4096, 1556112474.453495)]
# Add an item 10 seconds back
telstate.add('n_chans', 1024, ts=time.time() - 10)
# Attributes cannot be changed (only deleted)
telstate.add('no_change', 1234, immutable=True)
# Adding it again is OK as long as the value doesn't change
telstate.add('no_change', 1234, immutable=True)
# Simpler notation for setting attributes
telstate['no_change'] = 1234
# Will raise katsdptelstate.ImmutableKeyError
telstate['no_change'] = 456
# Create a new view with namespace 'ns' and standard underscore separator
view = telstate.view('ns')
# Insert a new attribute in this namespace and retrieve it
view['x'] = 1
print(view['x']) # -> 1
print(view.prefixes) # -> ('ns_', '')
print(view.keys()) # -> ['n_chans', 'no_change', 'ns_x']
There is also an interface that works with asyncio. Use katsdptelstate.aio.TelescopeState
instead of katsdptelstate.TelescopeState
. Functions that interact with the database are now coroutines. Python 3.6+ is required.
There are a few differences from the synchronous version, partly necessary due to the nature of asyncio and partly to streamline and modernise the code:
- The constructor only takes a backend, not an endpoint. See below for an example of how to construct a redis backend.
- There is currently no support for reading or writing RDB files; you'll need to create a synchronous telescope state client that connects to the same storage.
- There is no support for attribute-style access.
- Item-style access is supported for read (
await ts.get('key')
), but not for write. Useawait ts.set('key', 'value')
instead to set immutable keys. - Instead of
key in ts
, useawait ts.exists(key)
. - The
wait_key
andwait_indexed
methods do not take a timeout or a cancellation future. They can be used with asyncio's cancellation machinery. The async-timeout package is useful for timeouts. - The backend should be closed when no longer needed to avoid warnings.
from katsdptelstate.aio import TelescopeState
from katsdptelstate.aio.redis import RedisBackend
# Create a connection to localhost redis server
backend = await RedisBackend.from_url('redis://localhost')
ts = TelescopeState(backend)
# Store and retrieve some data
await ts.set('key', 'value')
print(await ts.get('key'))
# Close the connections (do not try to use ts after this)
ts.backend.close()
await ts.backend.wait_closed()