A Performance Metrics Domain Agent (PMDA) that exports event records
must effectively act an event multiplexer. Events consumed by the
PMDA may have to be forwarded on to any number of monitoring tools
(or "client contexts"). These tools may be requesting events at
different sampling intervals, and are very unlikely to request an
event at the exact moment it arrives at the PMDA, making some form of
event buffering and queueing scheme a necessity. Events must be held
by the PMDA until either all registered clients have been sent them,
or until a memory limit has been reached by the PMDA at which point
it must discard older events as new ones arrive.
The routines described here are designed to assist the PMDA developer
in managing both client contexts and queues of events at a high
level. These fit logically above lower level primitives, such as
those described in pmdaEventNewArray(3), and shield the average PMDA
from the details of directly building event record arrays for
individual client contexts.
The PMDA registers a new queue of events using either
pmdaEventNewQueue or pmdaEventNewActiveQueue. These are passed an
identifying name (for diagnostic purposes, and for subsequent lookup
by pmdaEventQueueLookup) and maxmem, an upper bound on the memory (in
bytes) that can be consumed by events in this queue, before beginning
to discard them (resulting in "missed" events for any client that has
not kept up). If a queue is dynamically allocated (such that the
PMDA may already have clients connected) the pmdaEventNewActiveQueue
interface should be used, with the additional numclients parameter
indicating the count of active client connections. The return is a
negative error code on failure, suitable for decoding by the
pmErrStr(3) routine. Any non-negative value indicates success, and
provides a handle suitable for passing into the other API routines.
For each new event received by the PMDA, the pmdaEventQueueAppend
routine should be called, placing that event into the queue
identified by handle. The event itself must be contained in the
passed in buffer, having bytes length. The timestamp associated with
the event (time at which the event occurred) is passed in via the
final tv parameter.
In the PMDAs specific implementation of its fetch callback, when
values for an event metric have been requested, the
pmdaEventQueueRecords routine should be used. It is passed the queue
handle and the avp pmAtomValue structure to fill with event records,
for the client making that fetch request (identified by the context
parameter). Finally, the PMDA must also pass in an event decoding
routine, which is responsible for decoding the fields of a single
event into the individual event parameters of that event. The data
parameter is an opaque cookie that can be used to pass situation-
specific information into each decoder invocation.
Under some situations it is useful for the PMDA to export state about
the queues under its control. The accessor routines -
pmdaEventQueueClients, pmdaEventQueueCounter, pmdaEventQueueBytes and
pmdaEventQueueMemory provide a mechanism for querying a queue by its
handle and filling in a pmAtomValue structure that the
pmdaFetchCallBack method should return.
This page is part of the PCP (Performance Co-Pilot) project.
Information about the project can be found at ⟨http://www.pcp.io/⟩.
If you have a bug report for this manual page, send it to
email@example.com. This page was obtained from the project's upstream
Git repository ⟨https://github.com/performancecopilot/pcp.git⟩ on
2020-06-09. (At that time, the date of the most recent commit that
was found in the repository was 2020-06-09.) If you discover any
rendering problems in this HTML version of the page, or you believe
there is a better or more up-to-date source for the page, or you have
corrections or improvements to the information in this COLOPHON
(which is not part of the original manual page), send a mail to
Performance Co-Pilot PCP PMDAEVENTQUEUE(3)