Single EPICS PVs#

In this tutorial we will read, write, and monitor an EPICS PV in ophyd.

Set up for tutorial#

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/ status
decay                            RUNNING   pid 2695, uptime 0:00:01
mini_beamline                    RUNNING   pid 2696, uptime 0:00:01
random_walk                      RUNNING   pid 2697, uptime 0:00:01
random_walk_horiz                RUNNING   pid 2698, uptime 0:00:01
random_walk_vert                 RUNNING   pid 2699, uptime 0:00:01
simple                           RUNNING   pid 2700, uptime 0:00:01
thermo_sim                       RUNNING   pid 2701, uptime 0:00:01
trigger_with_pc                  RUNNING   pid 2702, uptime 0:00:01

Connect to a PV from Ophyd#

Let’s connect to the PV random_walk:dt from Ophyd. We need two pieces of information:

  • The PV name, random_walk:dt.

  • A human-friendly name. This name is used to label the readings and will be used in any downstream data analysis or file-writing code. We might choose, for example, 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")

Use it with the Bluesky RunEngine#

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".


|   seq_num |       time | time_delta |


|         1 | 21:08:16.8 |          3 |


generator count ['0cc3f81f'] (scan num: 1)


|   seq_num |       time | time_delta |


|         1 | 21:08:16.9 |          0 |

|         2 | 21:08:16.9 |          0 |

|         3 | 21:08:16.9 |          1 |

|         4 | 21:08:16.9 |          3 |


generator list_scan ['90052872'] (scan num: 2)


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.


|   seq_num |       time | time_delta |          x |


|         1 | 21:08:17.1 |            |         -1 |

|         2 | 21:08:17.6 |            |         -1 |

|         3 | 21:08:18.1 |            |         -1 |

|         4 | 21:08:18.6 |            |         -1 |

|         5 | 21:08:19.1 |            |         -1 |


generator count ['f19fbf67'] (scan num: 3)


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"))


|   seq_num |       time |          x |


|         1 | 21:08:19.7 |         -1 |

|         2 | 21:08:20.0 |         -1 |

|         3 | 21:08:21.0 |         -1 |

|         4 | 21:08:22.0 |         -1 |


generator monitor_x_for ['d7df8444'] (scan num: 4)


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.

Use it directly#

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': {'value': 1.0, 'timestamp': 1682197697.094126}}


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': {'source': 'PV:random_walk:dt',
  'dtype': 'number',
  'shape': [],
  'units': '',
  'lower_ctrl_limit': 0.0,
  'upper_ctrl_limit': 0.0,
  'precision': 0}}


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...")
print("Moved to 5.")
Moving to 5...

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.


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):
   reading =

There are two problems with this counterexample.

  1. We might not know how often we need to check for updates.

  2. We often want to watch multiple signals with different update rates, and this pattern would quickly become messy.

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)

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.

deque([{'x': {'value': -0.8998594794613708, 'timestamp': 1682197703.034473}},
       {'x': {'value': -0.8816911707418178, 'timestamp': 1682197708.037917}}],

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.