In this tutorial we will read, write, and monitor an EPICS PV in ophyd.
We'll start our IOCs connected to simulated hardware.
One implements a random walk. It has
just two PVs. One is a tunable parameter, random_walk:dt
, the time between
steps. The other is random_walk:x
, the current position of the random
walker.
The IOCs may already be running in the background. Run this command to verify that they are running: it should produce output with STARTING or RUNNING on each line. In the event of a problem, edit this command to replace status
with restart all
and run again.
!supervisor/start_supervisor.sh status
Let's connect to the PV random_walk:dt
from Ophyd. We need two pieces of
information:
random_walk:dt
.time_delta
.from ophyd.signal import EpicsSignal
time_delta = EpicsSignal("random_walk:dt", name="time_delta")
It is conventional to name the Python variable on the left the same as the
value of name
, but not required. That is, this is conventional...
a = EpicsSignal("...", name="a")
...but all of these are also allowed.
a = EpicsSignal("...", name="b") # local variable different from name
a = EpicsSignal("...", name="some name with spaces in it")
a = b = EpicsSignal("...", name="b") # two local variables
Next let's connect to random_walk:x
. It happens that this PV is not
writable---any writes would be rejected by EPICS---so we should use a read-only
EpicsSignal, ophyd.signal.EPICSSignalRO
, to represent it in in ophyd. In
EPICS, you just have to "know" this about your hardware. Fortunately if, in our
ignorance, we used writable ophyd.signal.EpicsSignal
instead, we could
still use it to read the PV. It would just have a vestigial set()
method
that wouldn't work.
from ophyd.signal import EpicsSignalRO
x = EpicsSignalRO("random_walk:x", name="x")
time_delta.wait_for_connection()
x.wait_for_connection()
The signals can be used by the Bluesky RunEngine. Let's configure a RunEngine to print a table.
from bluesky import RunEngine
from bluesky.callbacks import LiveTable
RE = RunEngine()
token = RE.subscribe(LiveTable(["time_delta", "x"]))
Because time_delta
is writable, it can be scanned like a "motor". It can
also be read like a "detector". (In Bluesky, all things that are "motors" are
also "detectors".)
from bluesky.plans import count, list_scan
RE(count([time_delta])) # Use as a "detector".
RE(list_scan([], time_delta, [0.1, 0.3, 1, 3])) # Use as "motor".
For the following example, set time_delta
to 1
.
from bluesky.plan_stubs import mv
RE(mv(time_delta, 1))
We know that x
represents a time-dependent variable. We can "poll" it at
regular intervals
RE(count([x], num=5, delay=0.5)) # Read every 0.5 seconds.
but this required us to choose an update frequency (0.5
). It's often better
to rely on the control system to tell us when a new value is available. In
this example, we accumulate updates for x
whenever it changes.
from bluesky.plan_stubs import monitor, unmonitor, open_run, close_run, sleep
def monitor_x_for(duration, md=None):
yield from open_run(md) # optional metadata
yield from monitor(x, name="x_monitor")
yield from sleep(duration) # Wait for readings to accumulate.
yield from unmonitor(x)
yield from close_run()
RE.unsubscribe(token) # Remove the old table.
RE(monitor_x_for(3), LiveTable(["x"], stream_name="x_monitor"))
If you are a scientist aiming to use Ophyd with the Bluesky Run Engine, you may stop at this point or read on to learn more about how the Run Engine interacts with these signals. If you are a controls engineer, the details that follow are likely important to you.
Note: These methods should not be called inside a Bluesky plan.
The signal can be read. It return a dictionary with one item. The key is the
human-friendly name
we specified. The value is another dictionary,
containing the value
and the timestamp
of the reading from the control
system (in this case, EPICS).
time_delta.read()
Additional metadata is available. This always includes the data type, shape, and source (e.g. PV). It may also include units and other metadata.
time_delta.describe()
This signal is writable, so it can also be set.
time_delta.set(10).wait() # Set it to 10 and wait for it to get there.
Sometimes hardware gets stuck or does not do what it is told, and so it is good practice to put a timeout on how long you are willing to wait until deciding that there is an error that needs to be handled somehow.
time_delta.set(10).wait(timeout=1) # Set it to 10 and wait up to 1 second.
If the signal fails to arrive, a TimeoutError
will be raised.
Note that set(...)
starts the motion but does not wait for it to
complete. It is a fast, "non-blocking" operation. This enables you to run
code between starting a motion and completing it.
status = time_delta.set(5)
print("Moving to 5...")
status.wait(timeout=1)
print("Moved to 5.")
To move more than one signal in parallel, use the ophyd.status.wait
function.
from ophyd.status import wait
# Given signals a and b, set both in motion.
status1 = a.set(1)
status2 = b.set(1)
# Wait for both to complete.
wait(status1, status2, timeout=1)
What's the best way to read a signal that changes over time, like our x
signal?
First, set time_delta
to a reasonable value like 1
. This controls the
update rate of x
in our random walk simulation.
time_delta.set(1).wait()
We could poll the signal in a loop and collect N readings spaced T seconds apart.
import time
# Don't do this.
N = 5
T = 0.5
readings = []
for _ in range(N):
time.sleep(T)
reading = x.read()
readings.append(reading)
There are two problems with this counterexample.
Alternatively, we can use subscription.
from collections import deque
def accumulate(value, old_value, timestamp, **kwargs):
readings.append({"x": {"value": value, "timestamp": timestamp}})
readings = deque(maxlen=5)
x.subscribe(accumulate)
When the control system has a new reading
for us, it calls
readings.append(reading)
from a background thread. If we do other work or
sleep for awhile and then check back on readings
we'll see that it has some
items in it.
time.sleep(3)
readings
It will keep the last 5
. We used a collections.deque
instead of a
plain list
here because a list
would grow without bound and, if left to
run long enough, consume all available memory, crashing the program.