replace use of dict with orddict for per-msg_store partitions
which is much faster for small dicts

merge default into bug21673

add rabbit_misc:orddict_cons helper
analogous to dict_cons

Unit test, and fix a bug found by the test.

Spelling

Merge bug22992 into default.

refactor: extract rates into separate part of vq state

Rewrite the channel stats again to use ets. Not sure if this is any faster, maybe a bit.

Cosmetics

Merged default into bug21387

Merging default into bug 21673

Bump specs requirements to R14A

Increase priority of stats messages for channel and queue. Don't use gen_server2 for reader since it's not a real gen_server anyway.

Fixed channel shutdown mechanisms

Correct shutdown to use API

Store (channel, exchange) stats as well as (channel, exchange, queue) since we can't derive the former from the latter.

Store queue/exchange stats in the process dictionary since that's rather a lot faster than dict:

Well the rabbit_tests now pass owing to being able to support a very similar API to previously wrt channel, but shutdown seems to be sporadically successful at best.

Whoops - revert to avoid deadlock with supervisor trying to call itself during child creation

Invoking os:timestamp and timer:now_diff all the time is too expensive, use timers instead.

Add framing_channel:start_link/1 for use by Erlang client

Merging default into bug 15930

Merging default into bug 21922

use the new BQ:stop/0 function in tests
for better symmetry

Move logic for validating arguments to queue declaration out of the initialisation of the queue process.

If a new QPid appears more than once in the list, only monitor it once.

Store (channel, queue, exchange) stats instead of (channel, exchange). Unify the two stats dictionaries in the channel.

Rename some variables.

Standardise on "stats" rather than "statistics" since we were getting rather verbose. And inconsistent.

Store both channel-exchange stats and channel-queue stats in the same format.

simplified code

reformatted comments

list_connections can now print a peer's ssl certificate

rabbit_amqqueue:stop already stops the msg_store

merge default into bug21673

flesh out backing_queue API

cosmetic

cosmetic

more vq test refactoring

Refactorings in observation of commonality between queues and exchange arg equivalences. Also cosmetics, and various specs fixes.

Merge the timed/delayed restart code originally from supervisor3 in shovel into supervisor2

vq test refactoring

bug22917 merged into default.

default merged into bug22917

cosmetics

Merging bug 22677 onto default

reporting of cluster node types now works

Cherry-pick the last differences between bug 22889 and amqp_0_9_1.

Make more consistent with bug 22889 - see comment 83 on that bug

remove superfluous test
this is covered elsewhere already, notably in
test_variable_queue_partial_segments_delta_thing.

Merged default into bug 22889

Merge default into amqp_0_9_1

This function was merged wrong; fix.

Merge bug22889 into amqp_0_9_1

made one of the vq tests a little less obscure
...and don't rely on timing

rabbit_plugin_activator blows up on xref_undef warnings

initial break-for-some warnings support in plugin activator

applied qa notes; refactoring

cosmetic

cosmetic

All sorts of improvements to the queue lease code - simplifications and removing dead code

Missed out arguments in binding_created.

Lifecycle events for queues, exchanges and bindings.

minor refactoring of vq tests

The channel has a few statistics / state of its own too.

Initial commit of Ask's work

simplify qi tests

Use list comprehensions to build the stats proplists.

simplify qi tests

Switch to using proplists rather than records for events.

Make channel monitor queues for which it is gathering statistics, and remove the stats when the queue goes away.

simplify qi tests

cosmetic

simplify qi tests

refactor msg_store starting/stopping in tests
using the new funs in vq

flesh out vq API with funs useful for testing

merge default into bug21673

flesh out APIs somewhat
useful in tests

More documentation

That's not needed.

Improve documentation and document persistent vs transient msg store are for

Merging default into bug 15930 and a little debitrotting

Emit events for connection and channel statistics and creation / deletion.

refactor: simplify msg_store tests

refactor: simplify msg_store tests

refactor: simplify msg_store tests

tweak

refactor: take adantage of l/r symmetry in bpqueue tests

dict:find -> dict:fetch andalso lists:flatten -> lists:append

Merging bug 22987 into bug 21673

Cosmetics

optimise qi:sync call in tx commit
call it with acks for persistent & durable messages only

application:get_env/2 -> application:get_env/1

moved rabbitmq_cluster.config into the cluster_nodes field of rabbit in rabbitmq.config

refactor: 'cons_if' helper function

merge default into bug21673

cosmetic

tweak to assist testing
...outside the context of the rabbit app
cherry-picked from bug 21673.

'sync' -> 'idle_timeout'
cherry-picked from bug 21673.

tweak msg_store API to make it safer
...by not allowing clients to be deleted w/o terminating them too

tiny refactor

add a 'todo'

don't log warnings on clean startup of a msg_store

cleared up manpage

cosmetics

Use timer:now_diff, don't double-wrap.

q_pid -> qpid

Merged bug22140 into default.

Get queues to emit events for statistics.

don't forget to process transactional acks
...when the tx doesn't involve any persistent messages, thus plugging
a leak in pending_ack and qi.

To reproduce:

rr(amqp_connection).
Conn = amqp_connection:start_network(),
Ch = amqp_connection:open_channel(Conn),
Q = <<"Q">>,
amqp_channel:call(Ch, #'queue.declare'{queue = Q, durable = true, exclusive = true}),
amqp_channel:call(Ch, #'basic.publish'{routing_key = Q}, #amqp_msg{}),
amqp_channel:call(Ch, #'basic.get'{queue = Q}),
amqp_channel:call(Ch, #'tx.select'{}),
amqp_channel:call(Ch, #'basic.ack'{delivery_tag = 1}),
amqp_channel:call(Ch, #'tx.commit'{}),
ok.

and then

$ ./scripts/rabbitmqctl list_queues name messages_unacknowledged backing_queue_status
Listing queues ...
Q 0 [{q1,0},{q2,0},{delta,{delta,undefined,0,undefined}},{q3,0},{q4,0},{len,0},{pending_acks,1},{outstanding_txns,0},{target_ram_msg_count,infinity},{ram_msg_count,0},{ram_index_count,0},{avg_egress_rate,0.0},{avg_ingress_rate,0.0},{next_seq_id,1},{persistent_count,0}]
...done.

Note the pending_acks of 1.

add pending_acks to bq state report
it turns out to be useful

refactor: invert some conditionals for better readability

refactor: trim args for msg_store_callback
we only ever need a msg_store_callback when we have persisted pubs

another place were averages could end up being integers instead of floats

rate averages are meant to be floats

avoid >= between a number and an atom

improve type specs

cosmetic

merge default into bug21673

add some msg_status invariants
these are quite important for understanding much of the code

added exref option to systools:make_script

*WARNING*s are emitted when plugins reference undefined functions

cosmetic

renamed variables; alligned code

cosmetic

refactor

simplify remove_pending_ack
The pending_ack dict should always be empty at the end. Previously we
kept the ack records of persistent messages when KeepPersistent=true.

cosmetic

merging in from default

Merged heads

Changelog updates for release 1.8.1

Added tag rabbitmq_v1_8_1 for changeset 5d9e3ff32dcb

cosmetic: more sensible order

Merge bug22875 into default

>80 lines fix.

merge bug22875 into bug22964

added insightful comment

scope is not set via the -s flag

Merged bug22851 to junk

added coverage tests

added a flag to set_permissions to control their scope

set_permissions -check_user_named is the default and checks
permissions only for user named permissions. I.e. server generated
names are not checked.

set_permissions -check_all_resources enables the check for server
generated names.

I had to shorten the argument names in the man page because otherwise
the set_permissions line would have exceeded 80 characters.

All parameters passed from rabbit_control to rabbit_access_control are
binary(), hence why we're passing <<"check_user_name">> rather than
'check_user_named'.

Rabbit_access_control:set_permissions now takes 6 parameters. There's
also a 5 parameter version that sets the default value for Check and
calls the other one. I've added it because I don't want the default
value for Check in 10 different places.

bug22914 merged into default

replaced call bad:bad/0 with a call to exit/1

updated man page

fixed compile errors; all tests pass

permission tags are now displayed

check_all == 'true' => permissions are checked for server-generated names as well

added necessary fields to records

empty permissions are not printed as "" anymore

"" is changed to "^$" when checking permissions

% rabbitmqctl set_permissions guest "" "" ""
Setting permissions for user "guest" in vhost "/" ...
...done.

% rabbitmqctl list_permissions
Listing permissions in vhost "/" ...
guest "" "" ""
...done.

empty strings are displayed as ""

% rabbitmqctl list_permissions
Listing permissions in vhost "/" ...
guest "" "" ""
...done.

corrected regexp

Used character escaping because erlang-mode has problems correctly
parsing "^$".

simplifying refactor

Adjust the CONFIG_FILE default with the macports prefix

Remove IFS setting on the su command

The script has lots of IFS vulnerabilities before it reaches the su
command. If we really want to guard against IFS, we should set it at
the top. But how many shell scripts have you ever seen that do that?
If the user sets IFS, and breaks things, it seem reasonable to regard
that as their problem. There is no security issue here because you
should already be root or the rabbitmq user to run the script.

"make clean" should clean all the macports-generated files

"" is transformed to "$^" in rabbitmqctl set_permissions

Remove maybe_encode_properties

version should come first

export protocol/0 type
to avoid compiler warning

put the 'none' check back into ensure_content_encoded
it's safer and more obviously correct that way

refactor: extract error helper

don't log content on framing error
plus some minor refactors and cosmetic changes

slight optimisation

Merged bug22877 into default

finished transition from regexp to re

replaced regexp with re

0-9-1 error codes in a multi-prococol broker

Merge bug22886 into junk

This wiped out rabbit_framing_channel. Never mind.

Merge default into bug22889

Don't change the signature of ensure_content_decoded.

merged default into bug22877

Fix idiocy

Make sure one does not accidentally build a server containing old rabbit_framing when switching branches.

Store protocol with encoded properties, use it to transcode if needed.

deduplicate even the current node

added deduplication of cluster node list

Deduplication happens for both cluster and force_cluster because
clustering with the same node twice doesn't make sense in either case.

improved docs

refactor: introduce rabbit_misc:protocol_error/1
so that the channel can get away with knowing a little bit less about
error handling

Merging bug 22935 into default

amqp_error is a record. Duh.

Expose the persistent count in the vq status

Fix type of inner_fun, cosmetics.

Document ram_index_count

don't force storing of binary properties
see also bug 22957.

merge default into bug21673

merge heads

Update inner_fun spec.

Make add and delete more symmetrical, tidy up a bit.

Update comment.

Wrap much less in try / catch

remove incorrect comment
We actually never decode the properties on recovery (or later) since
the broker doesn't need them after the initial stages of message
processing.

minor refactor

cosmetic

Unwrap amqp_error and rethrow

merged bug22950 into default

Rename rabbit_reader_queue_collector to rabbit_queue_collector, and shorten some function and variable names.

merge heads

merge bug22896 into bug21673

Bump the msg_store file size limit when testing from 128bytes to 512bytes (normal operational limit is 16MB). This ensures that the GC code is hit