Age | Commit message (Collapse) | Author |
|
Communication between Erlang processes has conceptually always been
performed through asynchronous signaling. The runtime system
implementation has however previously preformed most operation
synchronously. In a system with only one true thread of execution, this
is not problematic (often the opposite). In a system with multiple threads
of execution (as current runtime system implementation with SMP support)
it becomes problematic. This since it often involves locking of structures
when updating them which in turn cause resource contention. Utilizing
true asynchronous communication often avoids these resource contention
issues.
The case that triggered this change was contention on the link lock due
to frequent updates of the monitor trees during communication with a
frequently used server. The signal order delivery guarantees of the
language makes it hard to change the implementation of only some signals
to use true asynchronous signaling. Therefore the implementations
of (almost) all signals have been changed.
Currently the following signals have been implemented as true
asynchronous signals:
- Message signals
- Exit signals
- Monitor signals
- Demonitor signals
- Monitor triggered signals (DOWN, CHANGE, etc)
- Link signals
- Unlink signals
- Group leader signals
All of the above already defined as asynchronous signals in the
language. The implementation of messages signals was quite
asynchronous to begin with, but had quite strict delivery constraints
due to the ordering guarantees of signals between a pair of processes.
The previously used message queue partitioned into two halves has been
replaced by a more general signal queue partitioned into three parts
that service all kinds of signals. More details regarding the signal
queue can be found in comments in the erl_proc_sig_queue.h file.
The monitor and link implementations have also been completely replaced
in order to fit the new asynchronous signaling implementation as good
as possible. More details regarding the new monitor and link
implementations can be found in the erl_monitor_link.h file.
|
|
There doesn't seem to be any science behind the long delays, and
the (newly introduced) dry run forces us to eat them twice, so
they've been shortened to more reasonable values.
|
|
|
|
#### Why do we need this new feature?
There are cases when a NIF needs to send a message, using `enif_send()`, to a long-lived process with a registered name.
A common use-case is logging, where asynchronous fire-and-forget messages are the norm.
There can also be cases where a yielding or dirty NIF or background thread may request a callback from a service with additional information it needs to complete its operation, yielding or waiting (with suitable timeouts, etc) until its state has been updated through the NIF module's API.
NIFs can only send messages to pids, and the lack of name resolution leaves a complicated dance between separate monitoring processes and the NIF as the only way to keep a NIF informed of the whereabouts of such long-lived processes.
Providing a reliable, built-in facility for NIFs to resolve process (or port) names simplifies these use cases considerably.
#### Risks or uncertain artifacts?
Testing has not exposed any significant risk.
The implementation behaves as expected on regular and dirty scheduler threads as well as non-scheduler threads.
By constraining the `enif_whereis_...()` functions to their minimal scopes and using patterns consistent with related functions, the implementation, testing, and maintenance burden is low.
The API and behavior of existing functions is unchanged.
#### How did you solve it?
While extending `enif_send()` to operate on a pid or an atom (as `erlang:send/2` does) was attractive, it would have entailed changing the type of its `to_pid` parameter and thereby breaking backward compatibility.
The same consideration applies to `enif_port_command()`.
That leaves a choice between 1, 2, or 3 new functions:
1. `enif_whereis()`
2. `enif_whereis_pid()` and `enif_whereis_port()`
3. All of the above.
While option (1), directly mimicking the behavior of `erlang:whereis/1`, is appealing, it poses potential problems if `pid()` or `port()` are subsequently implemented as non-integral types that must be bound to an owning `ErlNifEnv` instance.
Therefore, option (2) has been chosen to use `ErlNifPid`/`ErlNifPort` structures in the API to maintain proper term ownership semantics.
|
|
* maint:
Update copyright-year
Conflicts:
lib/dialyzer/src/dialyzer.hrl
lib/dialyzer/src/dialyzer_options.erl
lib/dialyzer/test/opaque_SUITE_data/src/recrec/dialyzer.hrl
lib/dialyzer/test/opaque_SUITE_data/src/recrec/dialyzer_races.erl
lib/hipe/icode/hipe_icode.erl
lib/hipe/main/hipe.erl
lib/hipe/main/hipe.hrl.src
lib/hipe/main/hipe_main.erl
|
|
|
|
Anywhere but the beam sources we shouldn't #include "erl_nif.h", because
what "erl_nif.h" does is: (1) fail to find it outside of -I dirs, (2)
then treat it as if it was written like <erl_nif.h>. Using <erl_nif.h>
skips (1).
More information can be found in 6.10.2 of the C standard.
Because the examples use "erl_nif.h", NIF projects in the Erlang
ecosystem copy this verbatim and make the same mistake.
|
|
|
|
|
|
* rickard/dirty_nif_SUITE-win-fix:
Fix windows
|
|
Dirty schedulers only execute NIFs, so having them execute the full
process_main function isn't necessary. Add dirty_process_main for
dirty schedulers to execute instead.
Add erts_pre_dirty_nif(), called when preparing to execute a dirty
nif.
Add more dirty NIF tests to verify that activities requiring the
process main lock can succeed when the process is executing a dirty
NIF.
|
|
|
|
|
|
|
|
|