Error message readability

Remove redundant code

Perform duplicate node check in a function

Cope with absence of epmd while performing pre-launch checks

Reimplement or manually merge all past changes for 23329 onto new head of default. In general the code's a bit cleaner due to better apis within msg_store generally

Merging bug23329 into junk (so bitrotted, better to start from scratch)

Demonstrate protocol_error.

Better diagnostics when authentication fails.

Merge default into bug 23554

Merge default into bug 23554 (epic debitrot)

Rip out old persister as it no longer works given the arrival of pub acks

Correction of specs.

Minor correction

add missing case

add stats for confirms

To recap, a published message is confirmed by the channel. A message
is confirmed only after all the queues it was published to confirm it.

With the current change, the emitted stats look like this:

{channel_exchange_stats,
[{{resource,<<"/">>,exchange,<<"direct">>},
[{confirm,545},{publish,545}]}]},
{channel_queue_exchange_stats,
[{{<0.204.0>,{resource,<<"/">>,exchange,<<"direct">>}},
[{confirm,545},{publish,545}]},
{{<0.195.0>,{resource,<<"/">>,exchange,<<"direct">>}},
[{confirm,545},{publish,545}]}]}]

The confirm field in channel_exchange_stats represents the number of
messages sent to that exchange that have also been confirmed. If the
exchanged routed the message to different queues, this number is only
increased when all queues have confirmed the message. If the message
was unroutable or was routed to 0 queues, this number is still
increased. This is the number of basic.confirms sent back to
publisher.

The confirm field in channel_queue_exchange_stats represents the
number of messages confirmed by that queue (but not necessarily
confirmed by the channel).

In channel_exchange_stats, if the number of confirms lags behind the
number of publishes, one of the queues is not confirming messages in a
timely fashion.

fix to compile and run under R12B5

merge default into bug23201

cosmetic

cosmetic

cosmetic: rename field/var

cosmetic

Fix

Consistency with other events.

Cosmetic

Fix

Oops.

Add consumer created / deleted events.

Consistent prelaunch checks on windows

Redid bug23056 changes atop newer base.

Running "hg rm".

Merged bug23056 into junk; I sure hope that doesn't break anything.

Add method name into the error tuple.

Simplify

Only look for AMQP errors in exits.

Remove quotes

More robust SASL PLAIN parsing - cope with any number of \0's being present without falling over.

Catch all types of exceptions, preserve the stack trace in the event that we have a non-AMQP error.

Merge heads

Merged bug23532 into default

Tone down the claims in the optimisation comment.

Merge from default

Merge bug23522 to junk.

Merged bug23565 into default

Added cosmetic change suggested in the bug report

fix to compile and run under R12B5

Merged bug23567 into default

simplify multiple confirm code

Merged bug23448 into default

Diagnostics on node start failure

Remove should_offer/1.

Rename plugin_activator to prelaunch

Check if nodename is available before startup

Merged bug20284 into default

cosmetic: put rabbit_net exports, specs, and funs in sensible order
and consistent too

experiment with uing {active,once} instead of prim_inet:async_recv

Exposing the publishing chpid right through to the BQ (and tidying up tests/types/specs)

Added tag rabbitmq_v2_2_0 for changeset c9a2e7f3f5e9

Changelogs for the 2.2.0 release

Make the tests work again. Also mirror_pids => extra_pids on the basis that it could be reusable for other purposes

Lesser of many evils...

Introduce some abstraction, reducing duplicate calls

Present the whole amqqueue record to the BQ (remove duplicated IsDurable flag, correct BQ callbacks)

Present the whole amqqueue record to the BQ

Added mechanism to dynamically choose the backing queue module. Ideally, this'll end up using the generic registry that's part of bug 23467 and allow plugins to register this stuff on boot

experimental single-frame publish/deliver commands

flesh out writer API for completeness

Introduce file_handle_cache:get_obtain_count and get_obtain_limit.

cosmetic

Merge with default

Cosmetic

merge with default

cosmetic

cosmetic

merging heads together

Merge bug23548 into default

cosmetic

bug23547 merged into default

Opps, they shouldn't have got there...

Opps, they shouldn't have got there...

Cope with files not existing (or just being empty)

cosmetic changes

No need to open these files at all exclusive. Which is an R14ism anyway.

Abstract out file locking, and revert to not using the R14-ism exclusive flag

Make sure we get the stacktrace out of a boot failure

Updated portfile as suggested by macports guru

Indentation

The resource is not a string.

let frame interleaving happen
but still optimise the delivery of short messages

merging heads of default

merge from default

change order of tuple elements

cosmetic

cosmetic

cosmetic

cosmetic + handle missing case

remove redundant record field

Dammit.

exit -> throw seeing as bug 23535 is invalid

Merging bug23519 into default (again)

Mnesia upgrade to convert user to internal_user records. Painful since the built-in Mnesia upgrade system does not let us change the type in a table.

Merge from bug23467 (and hence default)

Merge from default

Application of M-Q

Refactoring

gratuitous cosmetic

Merging bug23519 into default

Add helpful comment about windows flushing to rabbitmqctl

Slower rabbitmqctl shutdown on windows flushes output

Spec stuff.

Don't allow the backend to grant access to a vhost which does not otherwise exist!

Tiny simplifying refactor.

Merging bug 23335 into default

Merging bug 22692 into default

Merged bug23494 into default

It's rather crucial to call reduce_memory_use at the end of publish_delivered if we've added to the acks. Also reorder items in backing_queue_status so that they fit better on my 160-char wide xterms. It's also a bit more logical

Merging default into bug22692

cosmetic

merge bug23531 into default

Use internal_send_command, since we're sending to a 'raw' socket

Offer tx and non-tx exchange hooks

Merged bug23293 into default (again)

And, err, merge heads.

Merge amqp_0_9_1 into default.

Merge heads of amqp_0_9_1

re-merge bug21377
since it had a dangling head

merge heads

Allow Module:check_vhost_access and check_resource_access to return error tuples too.

Fix some whitespace issues, add a note

Merge bug23186 into default

plugin activator should use command line args instead of rabbit env vars

Merge bug23513 into default

Let an auth plugin specify which vhosts a user can see.

Allow auth backend modules to fail

Fix accidental renaming of an info item.

Improve consistency.

Merge bug23523 into default (rabbitmq-asroot-script-wrapper is pointless)

Not sure that's the right approach but at least a) EXTERNAL works via pluggable backends now and b) I've got it out of my head.

Increase the amount of correctness

Move HOSTNAME and NODENAME to rabbitmq-env

Remove OS-conditional and use hostname more conservatively

Be consistent about refused return

Merged default into bug23467

Umm, these are probably important

Rather impressionistic sketch of how pluggable auth backends might start out.

Make scripts work on Solaris

Remove untrue statement

Allow people to run commands as non-root/rabbitmq if they have reconfigured stuff to allow this to happen.

Remove rabbitmq-asroot-script-wrapper

Stop mnesia while taking the backup

Take a backup before upgrading.

Minor refactorings, and a couple of fixes

Merge add_internal_to_[exchange|durable_exchange] into single upgrade function.

Per Simon's feedback---they really belong together.

One head is better than two

Merging bug bug23498 to default

Merging bug23505 into default

Merged bug 23424 into default

merge heads

oops

minor optimisation: ask for next piece of data before processing
This reduces the likelihood of the reader process stalling,
i.e. ending up in 'receive' when there is nothing in the mailbox.

remove some cruft in reader, and some minor tweaks

Add Mnesia db migration for rabbit_durable_exchange table.

The broker should now be able to auto-upgrade both rabbit_exchange and
rabbit_durable_exchange tables at boot time.

Merge from default.

some routing optimisation
- handle the very common "message goes through single exchange" case
specially
- use a gb_set in all other cases - it performs much better than a set
for smaller sets, and not too badly for very large sets either.

merge from default

fix confirm after basic.reject{requeue=false}

refactoring

cosmetic: remove superfluous imports

Merge bug23460 (emit hibernate time in stats)

Throw rather than return errors

Merged default into bug23498

Added queue_ttl upgrade

Matthew points out that erlang:now() can be rewritten as now().

Merge bug23488 into default (Remove rabbit_misc:intersperse)

Reoprt idle_since pseudo-info-item.

Fix

Fixed windows scripts

Merging bug 23496 to default

Merge bug23498 heads

Don't open a security hole if server is misconfigured.

Specs.

Fix default config

After reading RFC5802 I'm not convinced this approach can be salvaged (in a sensible amount of time) so let's replace it with something that is a simple demo.

Merged from default

merge bug23495 into default

Fixed thinkos in calls to string:join.

Merged bug23493 into default

attempt to correct specs

Merge bug23354 (remove xmlto and half of universe as macports deps)

merge default into bug23498

cosmetic

make build_acyclic_graph return errors
and refine types

Replace catch with try; we want to show all kinds of boot errors in a readable way.

Remove the beginning of the per-mechanism options thing, in the interest of getting this thing finished.

merge default into bug23498

cosmetic

use {error,_} returns and throws instead of exits

Make sockets limit infinite on Windows.

Typo

Don't use records or record_info, explain why

Merging bug 23481 to default

exit->throw, *_or_exit -> ensure_*, and some cosmetics