Schemas

Lower bound of the device-instance range to scan. Requires high_limit.

Upper bound of the device-instance range to scan. Requires low_limit.

Target one specific device (directed Who-Is) instead of broadcasting - a local ip[:port] or a routed net:mac.

Broadcast on this one remote network (net:*) instead of globally.

false keeps the Who-Is to this subnet (a local broadcast), as before Baler#126.

Seconds to wait for I-Am responses. Defaults to bacpypes3's own Who-Is timeout (3s) if omitted.

The address the device responded from, as every other call takes it: a BACnet/IP address (no port when it's the standard 47808), or for a device behind a router net:mac - its network number and its MAC on that network (5:12 for MS/TP MAC 12 on network 5, 7:10.0.1.20 for a BACnet/IP device on a routed IP network).

The device's network number when it's behind a router; null on this subnet.

The device's MAC on its network (the IP address on this subnet).

Largest APDU the device accepts, from its I-Am (480 is typical of MS/TP).

The real BACnet object identifier - what create-a-connection needs, not a display name.

object-name if the device reports one, otherwise falls back to the object_identifier itself.

Current value at learn time - a snapshot, not a live feed. null if the object has none (e.g. a network-port) or the property read failed.

MQTT topic segment for this point - freely renameable from what /learn suggested.

The BACnet units /learn reported for this point (pass it through). Published as the point's Haystack unit tag.

The description /learn reported. Its words help tag the point (rules/bacnet_to_haystack_tags.yaml).

An explicit Haystack unit symbol for the point's unit tag, overriding units - for a device that reports none or the wrong one.

Seconds between reads of this specific point, overriding the connection-level default. Each point runs its own independent poll loop - a fast-changing point and a slow one on the same device don't have to share a cadence.

Enables change-of-value filtering for this point, matching Haystack's own hisCollectCov semantics: true publishes on any change; a number is a numeric tolerance the value must move by before it's published (non-numeric values always compare by equality regardless). Omit to publish every poll unconditionally (the default).

Never publish COV-triggered changes faster than this many seconds apart, even if the value keeps changing. Only meaningful alongside his_collect_cov.

Alternative to his_collect_cov - publish at most this often regardless of whether the value changed, independent of poll_interval (matches Haystack's hisCollectInterval).

Overrides the connection's subscribe_cov for this point - e.g. false for a point whose device floods notifications. (Not the same thing as his_collect_cov, which filters what gets published, however the value arrived.)

The controller's TrendLog object that logs this point (e.g. trend-log,3). After Windrow was down, or couldn't reach the device, the readings it missed are read back from that log with ReadRange and published on the point's topic with their own timestamps, not retained (so the live value stays the retained one). Goes back at most the connection's backfill_window.

What the schedule writes then: a number, active/inactive for a binary target, or null (relinquish).

The device's own timestamp: a date-time, a time of day, or #n for a sequence number.

When Windrow received it (unix seconds).

ip[:port] on this subnet, or net:mac for a device behind a router (5:12) - as /discover reports it. A broadcast address is a 400.

MQTT topic prefix every point on this connection publishes under, as {topic_prefix}/{label}.

Name of the device this connection polls - what it announces itself as (windrow.announce), or a DNS host name. When set, Windrow re-points the connection's address when the device announces (or resolves to) a new one. Empty clears it.

Seconds between poll cycles.

Read with ReadPropertyMultiple. Reads are batched per device, across every connection on it: points on one device tick together and go out as one request per read_multiple_max objects. A device that rejects RPM is read with one ReadProperty per point from then on; one whose answer doesn't fit an APDU gets smaller batches. false keeps one ReadProperty per point for this connection.

Most objects per ReadPropertyMultiple to this device. Default: what fits the device's max APDU (from its I-Am), at most WINDROW_RPM_MAX_POINTS (50). With several connections on one device the smallest value wins.

SubscribeCOV for every point (a point's own subscribe_cov overrides it), so a change publishes as soon as the device notifies. While a point's subscription is live it is still polled, every cov_poll_interval, in case a notification is lost; if the subscription fails it's polled at its normal interval and re-subscribed with back-off, and an object the device won't do COV for is just polled.

Seconds each COV subscription asks the device for. Renewed at half that.

Seconds between the safety-net polls of a point whose COV subscription is live.

Forward the device's intrinsic-reporting alarms (Baler#164). Windrow adds itself to the recipient-list of every notification-class object on the device (this changes the controller's configuration, hence opt-in), re-checks that every WINDROW_ALARM_REFRESH seconds (default 3600), and reads the active alarms at start. Each event notification is published retained on {topic_prefix}/{label}/alarm for a point of this connection, or {topic_prefix}/alarms/{type}-{instance} for another object, as the usual {point, value, ts} envelope whose value is the event (see AlarmEvent). Turning it off or deleting the connection removes Windrow's recipient entries again.

Most seconds of history a trend-log backfill reads back for this connection's points with trend_log; 0 turns backfill off.

Optional - the login this device's own UI/gateway needs, if any. Write-only on create; a created/listed Connection always shows this redacted (see RedactedCredential).

ip[:port] on this subnet, or net:mac for a device behind a router (5:12) - as /discover reports it. A broadcast address is a 400.

MQTT topic prefix every point on this connection publishes under, as {topic_prefix}/{label}.

Optional - the login this device's own UI/gateway needs, if any. Write-only on create; a created/listed Connection always shows this redacted (see RedactedCredential).

Listings only: whether this connection is actually reading, since the process started. Kept in memory and never stored on the record; it resets when the connection is updated. Times are unix seconds.

Name of the device this connection polls - what it announces itself as (windrow.announce), or a DNS host name. When set, Windrow re-points the connection's address when the device announces (or resolves to) a new one. Empty clears it.

Seconds between poll cycles.

Read with ReadPropertyMultiple. Reads are batched per device, across every connection on it: points on one device tick together and go out as one request per read_multiple_max objects. A device that rejects RPM is read with one ReadProperty per point from then on; one whose answer doesn't fit an APDU gets smaller batches. false keeps one ReadProperty per point for this connection.

Most objects per ReadPropertyMultiple to this device. Default: what fits the device's max APDU (from its I-Am), at most WINDROW_RPM_MAX_POINTS (50). With several connections on one device the smallest value wins.

SubscribeCOV for every point (a point's own subscribe_cov overrides it), so a change publishes as soon as the device notifies. While a point's subscription is live it is still polled, every cov_poll_interval, in case a notification is lost; if the subscription fails it's polled at its normal interval and re-subscribed with back-off, and an object the device won't do COV for is just polled.

Seconds each COV subscription asks the device for. Renewed at half that.

Seconds between the safety-net polls of a point whose COV subscription is live.

Forward the device's intrinsic-reporting alarms (Baler#164). Windrow adds itself to the recipient-list of every notification-class object on the device (this changes the controller's configuration, hence opt-in), re-checks that every WINDROW_ALARM_REFRESH seconds (default 3600), and reads the active alarms at start. Each event notification is published retained on {topic_prefix}/{label}/alarm for a point of this connection, or {topic_prefix}/alarms/{type}-{instance} for another object, as the usual {point, value, ts} envelope whose value is the event (see AlarmEvent). Turning it off or deleting the connection removes Windrow's recipient entries again.

Most seconds of history a trend-log backfill reads back for this connection's points with trend_log; 0 turns backfill off.

waiting = nothing polled yet; partial = some points failing; error = every point failing.

When a value was last published.

Labels of the points whose last read failed (at most 20).

unix seconds

modbus_discover only - hosts probed so far.

When done - what the synchronous route returns. Left out of the listing.

Modbus unit id.

Optional - a point in its stored object form (type, function, byte/word order...), as in a connection's points. Without it, register is read as one unsigned holding register.

The decoded value (a number or boolean).

Devices that have announced themselves (windrow.announce), ever.

Of those, how many announced within the last 3 intervals.

Announcements refused since start (outside --announce-allow, bad signature, malformed).

BACnet mode and, for a foreign device, its BBMD registration (see /bacnet/network).

BACnet requests sent since startup by kind

Subnets this node sits on (its BACnet --address and any --modbus-scan-cidr), as CIDRs.

Stored encrypted at rest (Fernet, keyed by the WINDROW_CREDENTIAL_KEY env var) - never returned in plaintext again after this call's response.

Shown once, here only - every later GET of the owning record redacts this to null.

Always null in any GET/list response - see CredentialResponse for the one-time-shown plaintext.

Subnet to sweep, in CIDR notation. No default - Modbus periodic/on-demand discovery only ever scans a subnet the operator explicitly supplied.

Max simultaneous TCP-connect attempts, bounded so a large sweep doesn't exhaust local ephemeral ports/file descriptors.

Seconds to wait for each host's TCP connect.

From Modbus function 0x2B/0x0E (Read Device Identification) - null if the device doesn't implement it, which is common and not an error.

0-based protocol address, not the 1-based "Modicon" number some register maps print: holding register 40001 is register: 0, input register 30010 is function: input, register: 9, coil 00001 is function: coil, register: 0. A 32-bit type spans register and register + 1.

Read with FC3 (holding), FC4 (input registers), FC1 (coils) or FC2 (discrete inputs).

Register types only.

32-bit types only. big = high word in the lower register (Modbus convention); little = word-swapped.

Byte order within each 16-bit register. little = byte-swapped.

Haystack unit, published as the point's unit tag. Don't also name this point in the connection's units.

Holding registers and coils only. Tags the point writable so Baler accepts a pointWrite to it, and lets POST /write / point_write command it (FC6 for one register, FC16 for two, FC5 for a coil).

MQTT topic prefix every point on this connection publishes under, as {topic_prefix}/{label}.

{label: point} - Modbus has no self-describing object list, so these are supplied directly from the vendor's register map. Each point is either a bare register address (the original form: one holding register read as an unsigned 16-bit integer) or a ModbusPoint object naming its function code, data type, word/byte order, scale/offset, unit and writability. Both forms can be mixed in one connection.

Name of the device this connection polls - what it announces itself as (windrow.announce), or a DNS host name. When set, Windrow re-points the connection's address when the device announces (or resolves to) a new one. Empty clears it.

Modbus unit/slave id.

Seconds between poll cycles - one shared interval for every point on this connection (unlike BACnet connections, Modbus connections don't support a per-point poll_interval).

Optional Haystack unit symbol per point label (a register is just a number). Published as each point's unit tag on {topic_prefix}/{label}/tags. A point object can carry its own unit instead; naming one point in both is a 400.

MQTT topic prefix every point on this connection publishes under, as {topic_prefix}/{label}.

{label: point} - Modbus has no self-describing object list, so these are supplied directly from the vendor's register map. Each point is either a bare register address (the original form: one holding register read as an unsigned 16-bit integer) or a ModbusPoint object naming its function code, data type, word/byte order, scale/offset, unit and writability. Both forms can be mixed in one connection.

null if no credential has been set for this device/connection.

Listings only: whether this connection is actually reading, since the process started. Kept in memory and never stored on the record; it resets when the connection is updated. Times are unix seconds.

Read-only. The priority levels commanded on each writable point through POST /write, as {label: {level: value}} (levels as strings). Modbus devices have no priority array, so Windrow keeps this one and writes the highest level's value. Absent when nothing is commanded.

Name of the device this connection polls - what it announces itself as (windrow.announce), or a DNS host name. When set, Windrow re-points the connection's address when the device announces (or resolves to) a new one. Empty clears it.

Modbus unit/slave id.

Seconds between poll cycles - one shared interval for every point on this connection (unlike BACnet connections, Modbus connections don't support a per-point poll_interval).

Optional Haystack unit symbol per point label (a register is just a number). Published as each point's unit tag on {topic_prefix}/{label}/tags. A point object can carry its own unit instead; naming one point in both is a 400.

Name of the device this connection polls - what it announces itself as (windrow.announce), or a DNS host name. When set, Windrow re-points the connection's address when the device announces (or resolves to) a new one. Empty clears it.

Re-point the connection (also done automatically for a named device that moves).

Replaces every point. Labels must be unique.

No +/#; a trailing / is dropped.

Read with ReadPropertyMultiple. Reads are batched per device, across every connection on it: points on one device tick together and go out as one request per read_multiple_max objects. A device that rejects RPM is read with one ReadProperty per point from then on; one whose answer doesn't fit an APDU gets smaller batches. false keeps one ReadProperty per point for this connection.

Most objects per ReadPropertyMultiple to this device. Default: what fits the device's max APDU (from its I-Am), at most WINDROW_RPM_MAX_POINTS (50). With several connections on one device the smallest value wins.

SubscribeCOV for every point (a point's own subscribe_cov overrides it), so a change publishes as soon as the device notifies. While a point's subscription is live it is still polled, every cov_poll_interval, in case a notification is lost; if the subscription fails it's polled at its normal interval and re-subscribed with back-off, and an object the device won't do COV for is just polled.

Seconds each COV subscription asks the device for. Renewed at half that.

Seconds between the safety-net polls of a point whose COV subscription is live.

Forward the device's intrinsic-reporting alarms (Baler#164). Windrow adds itself to the recipient-list of every notification-class object on the device (this changes the controller's configuration, hence opt-in), re-checks that every WINDROW_ALARM_REFRESH seconds (default 3600), and reads the active alarms at start. Each event notification is published retained on {topic_prefix}/{label}/alarm for a point of this connection, or {topic_prefix}/alarms/{type}-{instance} for another object, as the usual {point, value, ts} envelope whose value is the event (see AlarmEvent). Turning it off or deleting the connection removes Windrow's recipient entries again.

Most seconds of history a trend-log backfill reads back for this connection's points with trend_log; 0 turns backfill off.

Name of the device this connection polls - what it announces itself as (windrow.announce), or a DNS host name. When set, Windrow re-points the connection's address when the device announces (or resolves to) a new one. Empty clears it.

{label: point} - Modbus has no self-describing object list, so these are supplied directly from the vendor's register map. Each point is either a bare register address (the original form: one holding register read as an unsigned 16-bit integer) or a ModbusPoint object naming its function code, data type, word/byte order, scale/offset, unit and writability. Both forms can be mixed in one connection.

Replaces the {label: unit} map.

Unix epoch seconds.

Unix epoch seconds.

BACnet only.

BACnet only.

BACnet only.

Modbus only.

Modbus only.

Modbus only.

Modbus only.

Modbus only.

Modbus only, and only on a device added by address (POST /devices) - a sweep can't learn the unit id.

Set once a matching connection exists for this device.

null if no credential has been set for this device/connection.

Set if the best-effort default-credential check found one that worked. Always null for BACnet devices (no HTTP-login concept applies - see /devices's description).

Human-readable explanation of what went wrong.