Connector Tuning

Overview • Settings • Conn Tuning Tags • ConnTuning • Details • Tracing • Poll Buckets

Overview

Connector tuning is the most important mechanism at your disposal to manage system and network performance. This chapter covers the tools for debugging and tuning connectors:

Settings

All connectors support the following library level tuning options via settings (tags on the library ext rec itself):

maxThreads

The connector framework is built using Fantom's actor APIs. Each connector is itself an actor which executes on threads from a shared thread pool. By default each connector library is allocated a max thread pool size of 100. This means that at most 100 connectors of one type can concurrently do processing. If you have more than 100 connectors, this potentially could starve connectors from access to required processing. Applying the maxThreads tag to your library's ext rec enables tuning of the thread pool size. If you wish all your connectors to have concurrent threads, then tune this value to match your number of connectors. However, sometimes it's prudent to balance the thread pool size to avoid excessive CPU and memory usage because threads have significant overhead.

Conn Tuning Tags

All connector types support a default number of tuning options on each connector using the following tags:

Each of the tags is discusssed in more detail below.

actorTimeout

The actorTimeout tag configures a duration for message timeouts. This determines how long to block on a synchronous message to the actor when the connector is busy blocking on I/O or processing previous messages. Its most typically used by the Conn.sendSync method used by many libs for connector specific functionality such as haystackReadAll().

This timeout is also used for all socket timeouts for connectors which use TCP (including all HTTP based connectors). Specifically this timeout is used to configure the connectTimeout and receiveTimeout for the underlying TCP sockets.

The timeout defaults to 1min if not explicitly configured.

connLinger

The connLinger tag configures a duration to linger open a connection. For many connectors the open process can be expensive. For example many HTTP based protocols require several requests to authenticate and acquire a session. So depending on access patterns, a connector could potentially thrash between the open and close state. Lingering alleviates this situation by keeping the connection open a little while before transitioning to the close state. For example, if you configure the connLinger timeout to be 1min and then ping the connector, then the connector will be held open for a minute before transitioning back to the close state. The default is 30sec if not configured.

connOpenRetryFreq

The connOpenRetryFreq tag configurations a duration between open retries. Some applications such as a watch or MQTT subscription require a connector to be pinned open. If the connector fails to open, then the system will automatically retry to open the connection periodically using this frequency. If not configured the default is 10sec.

connPingFreq

The connPingFreq tag configurations a duration between forced pings. The default behavior for connectors is to stay closed unless an application explicitly forces them open. If this tag is added to the connector then the framework will automatically force a periodic ping. It should be used when you wish to monitor the status of connectivity, but don't have any watched points to keep it open.

ConnTuning

The primary mechanism to tune point behavior is via connTuning configuration records. A connTuning record is used to store the following point level tuning options:

There are also some connector specific options such as:

Every connector point is assigned to exactly one tuning configuration via the connTuningRef tag. This tag is searched in the following order:

  1. Point rec itself
  2. Point's connector rec
  3. Connector's ext rec
  4. Fallback to default tuning for connector type

Example:

id: @fast-poll
dis: "Fast Polling"
connTuning
pollTime: 1sec

id: @slow-poll
dis: "Slow Polling"
connTuning
pollTime: 30sec

id: @conn
dis:"Modbus Connector"
conn
modbusConn
connTuningRef: @slow-poll

dis:"Fast Poll Point"
point
modbusPoint
modbusConnRef: @conn
modbusCur: "reg1"
connTuningRef: @fast-poll

dis:"Slow Poll Point"
point
modbusConnRef
modbusPointRef: @conn
modbusCur: "reg2"

In the example above we have two different tuning records. The Fast Poll Point has an explicit connTuningRef to the fast poll configuration. The Slow Poll Point inherits the connector's connTuningRef. In your own projects you can debug the assigned connTuning with the details.

The following sections provide additional details on each of these point level tuning tags.

pollTime

The pollTime tag specifies a duration Number which is the frequency used to poll a point for curVal. This tag is only used for connectors which use the buckets polling mode. Connectors which use a COV subscription model will ignore this value. If unspecified the default is 10sec.

staleTime

The staleTime tag specifies a duration Number used to transition a point's curStatus tag from "ok" to "stale". It ensures that users and applications are aware that data might not be fresh. The transition to stale occurs when all the following conditions are met:

  1. the point's curStatus is currently "ok"
  2. the point is not in a watch
  3. the last successful read exceeds the stale time

Note that we assume points in a watch are currently up-to-date even if their last read time exceeds the stale time. This is because change of value subscriptions might not be calling updateCurOk continuously if no changes are received. If unspecified the default is 5min.

writeMinTime

The writeMinTime tag specifies a duration Number used to throttle the frequency of writes to the remote device. For example if configured to 5sec, then writes will be issued no faster than 5sec. After a successful write occurs, if any writes are attempted within that 5sec window then they are queued as a pending write. After 5sec has elapsed the last pending write is issued to the connector's onWrite callback. Note that writeMinTime is only enforced after successful writes. If the connector reports a write failure, then writeMinTime is not enforced on subsequent attempts.

writeMaxTime

The writeMaxTime tag specifies a duration Number used to issue periodic rewrites to the remote device. For example if configured to 10min, then if no successful writes have been issued after 10min then a write is automatically scheduled to the connector's onWrite callback. The writeMaxTime does not go into effect until after the project reaches steady state.

writeOnOpen

The writeOnOpen marker tag is applied to issue a write whenever the connector transitions from closed to open. This policy is typically used when the remote device stores writes in RAM only and needs to be re-written after reboots.

writeOnStart

The writeOnStart marker tag is applied to issue a write when the system starts up. If omitted then the system suppresses the initial priority array evaluation.

Details

Both connector and connector points provide a plain text report for debugging we call details. You can query the details using the connDetails() function. In SkySpark you can view the details using the context menu or the Connectors view.

Connector Details

The connector details report contains the following sections:

  • Summary and configuration
  • Polling mode and polling bucket allocations
  • State variables managed by the framework (linger, poll times, etc)
  • Transient tags managed by the framework
  • Connector specific extra debug
  • Current actor message and threading details

Point Details

The point details report contains the following sections:

  • Summary and configuration
  • Addressing information for fooCur, fooWrite, fooHis
  • Transient tags managed by the framework
  • Connector specific extra debug
  • Watches on the point
  • Current value state
  • Write state
  • History sync state
  • History collection state
  • Writable priority array

Only the applicable sections are included.

Tracing

Connectors provide debug tracing which includes:

  • all dispatch actor messages
  • lifecycle transitions for open, ping, close
  • poll requests
  • connector specific debug for networking messages

In SkySpark the Connector|Trace view provides a tool to debug a connector's trace log. You can also manage tracing with these Axon functions:

Be aware that the trace log is stored in memory using a circular buffer. For connectors which trace their message payloads this can result in significant RAM usage. Tracing is always disabled by default on startup and must be manually enabled on a per connector basis. Use the connTraceDisableAll() function after a debugging session.

Poll Buckets

Most simple connectors support curVal synchronization via polling. When a connector's pollMode is defined as buckets then the framework automatically allocates all points to polling buckets. Polling buckets can be used to tune the grouping and frequency of polls. Points are grouped into buckets via their unique connTuningRef tag. Poll frequency is configured via the pollTime tag. Note that two different connTuning recs with the same pollTime are still modeled as two different buckets.

Example:

id: @tuning-fast-1
dis: "Fast Bucket 1"
connTuning
pollTime: 1sec

id: @tuning-fast-2
dis: "Fast Bucket 2"
connTuning
pollTime: 1sec

id: @tuning-slow
dis: "Slow Bucket"
connTuning
pollTime: 15sec

dis: "Point F1-A", connTuningRef: @tuning-fast-1
dis: "Point F1-A", connTuningRef: @tuning-fast-1
dis: "Point F2-A", connTuningRef: @tuning-fast-2
dis: "Point F2-B", connTuningRef: @tuning-fast-2
dis: "Point S-A", connTuningRef: @tuning-slow
dis: "Point S-B", connTuningRef: @tuning-slow

The configuration above will result in the following three poll buckets

Fast Bucket 1 contains F1-A, F1-B
Fast Bucket 2 contains F2-A, F2-B
Slow Bucket contains S-A, S-B

Buckets are automatically staggered at startup so that groups are not polled at the exact same interval. In the example above, we have two buckets configured with a 1sec pollTime. The system will automatically stagger the poll interval randomly within that 1sec window.

Under the covers a bucket is polled with the onPollBucket callback. However, not all protocols support a batch read. So it is possible that bucket polls might still require individual point level read requests.

Haxall 3.1.10 ∙ 28-Jun-2024 15:01 EDT