fq is a brokered message queue using a publish subscribe model. It is architected for performance and isn't (today) designed for large numbers of connected clients.
+------------+ +-----------+
|- exchange -|<-- (msg publication) --|- client0 -|
+------------+ +-----------+
|- routemap -|
+------------+
| | +---------+
| +--------------|- queue -|
| +---------+
+---------+ |
|- queue -| | +-----------+
+---------+ +---|- client1 -|
| +-----------+
| +-----------+
+--|- client2 -|
| +-----------+
|
+-----------+
|- client3 -|
+-----------+
The fqd process. The daemon through which all knowledge passes.
Peers are connected fqd processes. It is important to note that peers are unidirectional. If A peers with B, then A will act as a client to B. If you want bidirectional peering, you must specify that A peers with B and B peers with A. The system aims to prevent cyclic delivery of messages efficiently.
Adding peers is done directly via fqd's sqlite DB store:
; sqlite3 /var/lib/fq/fqd.sqlite
sqlite> INSERT INTO "upstream"
(host, port, source, password, exchange, program, permanent_binding)
VALUES('peerB',8765,'fqd-peera//mem:drop,private,backlog=4096','none','logging','prefix:"http.access.json."','false');
- C client - libfq
- Java client - fq.jar
- Node.js client - fq
- Go client - fq
- submission-only /submit API (see below)
A client is an applications connection to fq over TCP/IP to send or receive messages. A client makes two TCP/IP connections to fq. An application can present itself to fq as multiple clients at one time (by opening new pairs of connections). See Queues for reasons why.
Exchanges are like buses on which messages may be sent. You cannot send a message without doing so on an exchange. Exchanges are created within fq on-demand.
Queues are queues. If you stick something in one end, you should expect it to come out the other. A single queue may have multiple clients subscribed. When a client connects, it is attached to one and only one queue. If an application wishes to attach to more than one queue, it should present as multiple clients. Queues use a competitive consumption model meaning that if multiple clients are attached to a single queue, the messages sent to that queue will be distributed over the clients such that no two clients will see the same message.
Queues can be of type mem or disk. The contents of memory queues will not survive restarts.
Various parameters can be set on a queue using the syntax type:param1,param2.
Queues with the public parameter can have multiple clients connected to them (in which case they compete for messages). If you want a private queue you can specify the private parameter.
Queues can either have a block or drop policy. The drop policy means that messages that would be routed to a queue that is full will be dropped and never delivered. The block policy will cause the publisher to wait until there is room in the queue. The block policy makes no sense on a disk queue.
The backlog=<number> parameter will specify how many messages may be held in the queue before the block or drop policies are applied.
If you want a queue to be remembered by fqd, you can specify permanent as a flag. If you'd like for fqd to forget the queue after all clients have disconnected, you can specify the transient flag. If neither flag is specified, then an existing queue will retain its previous permanence setting or a new transient queue will be created.
A queue called bob will be in memory, allowed to have multiple clients connected to it, with a drop policy and an allowable message backlog of 100000 messages: bob/mem:public,drop,backlog=100000
A connection client will specify username/queue. A user "USER" connecting to the aforementioned queue would connect as USER/bob/mem:public,drop,backlog=100000
Messages are, of course, a payload and metadata.
Some are set by the broker.
- sender [set by the broker]
- hops (a list of fqd via which the message passed)
Others are set by the sender.
- exchange (up to 127 bytes)
- route (up to 127 bytes)
- id (128 bits). The first 64 bits the sender shall control, the latter 64bits the broker might control.
Routes and programs define how messages sent on exchanges are placed in queues:
- A receiver that connects to an fq-broker specifies a program that filters the messages on the exchange.
- A sender specifies a route for every message as part of the metadata
Programs follow the following syntax (cf. fqd.h):
PROGRAM: <prefix|exact>:string RULES*
RULE: (RULE)
RULE: (RULE && RULE)
RULE: (RULE || RULE)
RULE: EXPR
EXPR: function(args)
args: arg
args: arg, args
arg: "string"
arg: true|false
arg: [0-9][0-9]*(?:.[0-9]*)
functions are dynamically loadable with type signature
strings: s, booleans: b, integers: d
function: substr_eq(9.3,10,"tailorings",true)
C symbol: fqd_route_prog__substr_eq__ddsb(int nargs, valnode_t *args);
In particular:
- Every program starts with either
prefix:orexact: - The program
prefix:matches all rules - The program string is matched against the message route
The following rule functions are defined in fq_prog.c:
fqd_route_prog__sample__d()-- subsample the streamfqd_route_prog__route_contains__s()-- check if route contains a stringfqd_route_prog__payload_prefix__s()-- check if payload starts with prefixfqd_route_prog__payload_contains__s()-- check if payload contains a stringfqd_route_prog__true__()-- always true
Examples:
prefix:-- matches all messagesprefix:blaorprefix:"bla"-- matches all messages with rules starting with the sting 'bla'prefix: payload_prefix("M")-- matches messages where the payload starts with 'M'prefix:foo (payload_prefix("M") && route_contains("bar"))-- matches messages where the payload starts with 'M' and route starts with "foo" and moreover contains "bar"
Information on command and message protocol is found in docs/fq_protocol.md
The Fq protocol also acts as a non-compliant HTTP server (though compliant enough of most clients and browsers). Fq ships with a web UI that allows inspecting real-time state and performance.
exposes current exchange, queue, and client information.
An endpoint allowing message submission without a full and stateful Fq connection. It expects the following headers:
X-Fq-User,X-Fq-Route, andX-Fq-Exchange.
The HTTP client MUST provide a Content-Length header corresponding to the payload content (no chunked submission). The payload is treated as the raw message box without any special encoding.
Example:
curl -X POST -H "X-Fq-User: user" -H 'X-Fq-Route: bla' -H 'X-Fq-Exchange: test' localhost:8765/submit --data "TEST"
Requirements:
Generally:
make
make install
To build without libbcd support:
NO_BCD=1 make
FQ can be run in debug mode from the command line.
To run FQ in debug mode, kill any and all existing FQ processes, then enter the following command:
fq -g fq FQ_DEBUG=<flag values> <path to fqd>/fqd -D -c <path to fqd.sqlite>/fqd.sqlite -p <port number>
Flag values determine debug output type and can have the following values:
FQ_DEBUG_MEM = 0x00000001,
FQ_DEBUG_MSG = 0x00000002,
FQ_DEBUG_ROUTE = 0x00000004,
FQ_DEBUG_IO = 0x00000008,
FQ_DEBUG_CONN = 0x00000010,
FQ_DEBUG_CONFIG = 0x00000020,
FQ_DEBUG = 0x00000040,
FQ_DEBUG_PEER = 0x00000080,
FQ_DEBUG_HTTP = 0x00000100,
FQ_DEBUG_PANIC = 0x40000000
To debug more than one flag, simply OR the flag values. For example, to output
connection, configuration, and route information, set FQ_DEBUG equal to
0x00000034 (FQ_DEBUG_CONFIG|FQ_DEBUG_CONN|FQ_DEBUG_ROUTE).
For example, you can run FQ in debug mode with the variables shown below to output configuration, connection, and route information to the console:
fq -g fq FQ_DEBUG=0x00000034 /opt/circonus/sbin/fqd -D -c /opt/circonus/var/lib/fq/fqd.sqlite -p 8765