Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
branch: master
Fetching contributors…

Cannot retrieve contributors at this time

160 lines (126 sloc) 4.942 kb

pzq

A very simple store and forward device using ZeroMQ. It has some optional cluster capabilities in order to provide high availability of all accepted messages.

This version is only compatible with zeromq version 3.2 and above.

Building

$ mkdir build
$ cd build
$ cmake .. -DZEROMQ_ROOT=/path/to -DKYOTOCABINET_ROOT=/path/to
$ make

Options

Command-line options:
  --help                                produce help message
  --database arg (=/tmp/sink.kch)       Database sink file location
  --ack-timeout arg (=5000000)          How long to wait for ACK before 
                                        resending message (microseconds)
  --reaper-frequency arg (=2500000)     How often to clean up expired messages 
                                        (microseconds)
  --hard-sync                           If enabled the data is flushed to disk 
                                        on every sync
  --inflight-size arg (=31457280)       Maximum size in bytes for the in-flight
                                        messages database. Full database causes
                                        LRU collection
  --receive-dsn arg (=tcp://*:11131)    The DSN for the receive socket
  --send-dsn arg (=tcp://*:11132)       The DSN for the backend client 
                                        communication socket
  --monitor-dsn arg (=ipc:///tmp/pzq-monitor)
                                        The DSN for the monitoring socket
  --replicas arg (=0)                   Number of replicas that should
                                        created before acknowledging message to 
                    producer
  --nodes arg                           List of DSN for other cluster nodes 
                                        separated by a ','
  --timeout-nodes arg (=10000000)       How long to wait before considering a 
                                        node down (microseconds)
  --broadcast-dsn arg                   DSN used by other cluster nodes to 
                                        broadcast msessages.
                                        Broardcast port 
                                        should be the same for all nodes
  --timeout_replication arg (=100000)   How long to wait for replication before
                                        acknowledging producer with a 
                    replication error message

Consistency

--ack-timeout Defines how long to wait for an ACK for message delivered before scheduling it for retransmission.

--hard-sync Define this option for physical synchronization with the device, or leave out for logical synchronization with the file system.

--inflight-size Defines the maximum size in bytes for the messages that are in flight. Setting this database small can harm performance as LRU needs to run more often and the messages that were in flight need to be retransmitted.

Centos Notes

Older versions will require the boost141-devel package, available from EPEL.

Specify the location by building with

$ cmake .. -DBOOST_INCLUDEDIR=/usr/include/boost141 -DBOOST_LIBRARYDIR=/usr/lib64/boost141

Communication

In general, anything before following the empty message part should be considered protocol data. The client should assume that in future more header information can be added to the messages.

  • Producing a message
    +--------------------+
    | message id         |
    +--------------------+
    | 0 size part        |
    +--------------------+
    | 1..N message parts |
    +--------------------+
  • Producer ACK message
    +---------------------+
    | message id          |
    +---------------------+
    | status code 0/1     |
    +---------------------+
    | 0 size part         |
    +---------------------+
    | 0..1 status message |
    +---------------------+

Note: Status code 1 for success and 0 for failure. Status message is set in case of failure.

Note: In the case of a message successully stored in queue but all replications could not be created before timeout_replication, the status will be 1 but the status message will contain REPLICATION_FAILED

  • Consumer message
    +---------------------+
    | peer id             |
    +---------------------+
    | message id          |
    +---------------------+
    | sent time           |
    +---------------------+
    | ack timeout         |
    +---------------------+
    | 0 size part         |
    +---------------------+
    | 1..N message parts  |
    +---------------------+
  • Consumer ACK message
    +---------------------+
    | peer id             |
    +---------------------+
    | message id          |
    +---------------------+
    | status code 0/1     |
    +---------------------+      

Note: Status code 1 for success and 0 for failure.

TODO

  • Add tests
Jump to Line
Something went wrong with that request. Please try again.