| Age | Commit message (Collapse) | Author |
|---|
|
* rickard/make-fixes-17/OTP-15657:
Remove own configured RM make variable
|
|
Instead rely on gnu make's pre-defined RM variable which should
equal 'rm -f'
|
|
Spread a server over three nodes, one of which terminates a peer
connection, the other two to handle requests. Terminate transport on one
of the server nodes and ensure that answers come only from the other
two.
|
|
If the Session-Id optional value to node() mapping fails then hash
Session-Id to a node by default, instead of selecting the local node as
in the parent commit. The previous behaviour is configurable by setting
default = local in an options map.
Nodes make themselves part of the pool from which nodes are selected by
calling diameter_dist:attach/1 with the list of service names they are
willing to handle requests for, the local node being selected in the
absence of any attached nodes. The original idea was to base the node
pool on share_peers and/or use_shared_peers configuration, but that
configuration determines where outgoing requests can be sent, while
route_session/2 deals with incoming requests, so it's not obvious that
conflating the two is a good thing. (Also because
share_peers/use_shared_peers can be used in different ways; the former
could have been skipped entirely.)
The hashing effectively places nodes on a circle, a hashed Session-Id
being mapped to the nearest predecessor node (clockwise). Nodes are
rehashed with each Session-Id (with the id as salt) for a more even
distribution, at the cost of performance, although how high the cost or
how even the distribution has yet to be tested. Obviously, the larger
the number of attached nodes, the higher the cost. Adding/removing an
attached node only affects session ids that hash in the interval between
the added/removed node and its successor (hence consistent hashing).
Options are tweaked slightly compared to the parent commit, and it is
now possible to restrict the optional value mapping to specific Diameter
identities, to avoid mapping an id that was generated at the peer when
the peer is also implemented with the diameter application.
Note that diameter_dist is not yet an officially documented interface,
so could change. Documentation is in the module itself.
|
|
To be able to restrict how many AVPs will be examined (from the front of
a message) when looking for Session-Id, and to decide what to do with if
the AVP isn't found. Options are specified as a map of the following
form.
#{search => non_neg_integer(),
default => discard | mfa(),
dispatch => list() | mfa()}
The search member says how many AVPs to examine at most, from the front
of the message. If the optional value of a Session-Id is not the name of
a connected node then the default member determines what to do with the
request, handle it locally (the default), discard it, or invoke an MFA
on the Session-Id | false (if none was found) and diameter_packet record
to return a node() | false; if the latter then the request is discarded.
If a node is identified then the dispatch MFA is invoked on the node and
the request MFA (as three arguments), a list Opts being equivalent to
the MFA {erlang, spawn_opt, [Opts]}, and the default being the empty
list.
Integer- or list-valued options are equivalent to the corresponding map
with a single value.
Limiting the search is to avoid searching messages containing many AVPs
for a Session-Id that is known to occur near the header, since section
8.8 of RFC 6733 says this:
When present, the Session-Id SHOULD appear immediately
following the Diameter header (see Section 3).
There's no guarantee, but in practice it may well be known that peers
are respecting the RFC, and in that case limiting the search is a
defense against searching messages from a malicious peer unnecessarily.
The search is unlimited by default.
A default is only used when a search fails to locate a Session-Id, and
can be to discard the message, or have a node() or false be returned
from an MFA applied to the diameter_packet in question. The local node
is chosen by default.
|
|
That is, of functions that can be configured as spawn_opt MFAs in
transport configuration.
This commits adds the spawn_local described in the parent commit, and a
route_session that assumes that the local node initiates all sessions
with Session-Id returned by diameter:session_id/1, and handles incoming
requests on the node on which the id in question was returned,
diameter:session_id/1 using node() as optional value in the Session-Id
format.
|
|
The possibility of configuring an MFA as spawn_opt was added in commit
fd285079, the callback being passed an arity-0 fun to be applied in an
appropriate handler process. Replace the fun by a tuple to be passed to
diameter_traffic:request/1, to avoid passing funs between nodes when
handler processes are remote.
A list-valued spawn_opt is now equivalent to the following configured as
{spawn_opt, {Mod, spawn_local, [Opts]}}.
spawn_local(ReqT, Opts) ->
spawn_opt(diameter_traffic, request, [ReqT], Opts).
ReqT is passed by diameter and contains information that the callback
may want to decide where to handle the request in question (which wasn't
accessible with a fun), but this information isn't exposed in a
documented way. The intention is instead to add an own callback
implementation to make use of the information.
Note that application lookup now takes place in the watchdog process in
both the list-valued (or no configuration) and mfa-valued cases. Whether
this is good, bad, or (probably) inconsequential remains to be seen.
|
|
This is the {diameter, ack} message from a transport process to its
parent that requests that every Diameter request passed to the parent be
matched by a incoming send message, the atom false replacing a message()
in the case that the request is unanswered, allowing the transport to
keep track of the number of outstanding requests.
These were added in commit ca09cf7b, and are used to implement the
message_cb config in diameter_tcp/sctp, documented in commit cefcaa5c.
This commit documents the interface, to make it available to other
transport implementations.
|
|
* maint:
Updated OTP version
Prepare release
|
|
* maint-21:
Updated OTP version
Prepare release
|
|
* maint:
kernel runtime dependency to erts
erts: Add yield via timeout to inet read_packet
erts: Don't increase buffer when sctp sndbuf is set
erts: Only change inet buffer if not set
|
|
into maint
* lukas/erts/fix_inet_buffer_auto_adjust/OTP-15651/OTP-15652:
kernel runtime dependency to erts
erts: Add yield via timeout to inet read_packet
erts: Don't increase buffer when sctp sndbuf is set
erts: Only change inet buffer if not set
|
|
* rickard/bootstrap-build-fix:
Fix bootstrap build of erl_interface
|
|
|
|
|
|
|
|
|
|
Let sys.h define HAVE_OVERFLOW_CHECK_BUILTINS if the compiler supports
__builtin_mul_overflow() and the other overflow checking builtins.
The test is intentionally made in a sys.h and not as a configure test.
On Windows, beam_emu.c is always compiled using gcc, but the other
files are usually compiled with Microsoft's C compiler. With the
test in the header file, HAVE_OVERFLOW_CHECK_BUILTINS will be defined
when compiling beam_emu.c.
|
|
Starting in OTP 19 (in commit 9504c0dd71d0), the compiler emits
a test_unit instruction instead of a skip instruction at the end
of binary. We can do the same replacement in the loader to get
rid of the i_bs_skip_bits_all2 instruction.
|
|
On a 64-bit architecture, the size of any binary that would fit in the
memory must fit in a small, so we can fail immediately if the size
term is not a small.
|
|
The new compiler required adding support for Y register for all
binary matching instructions. That was (intentionally) done in a
naive way that simplicated duplicated the entire body of each
instruction.
Now it's time to be less naive. Rewrite the binary matching
instructions using micro instructions. Because some of the binary
instructions are huge, that will significantly decrease the size of
process_main().
When compiling with clang, a huge process_main() would mess up
profile-guide optimization resulting in a significant performance
degradation. On my Mac, profile-guide optimzation would decrease
the estone benchmark by 100K estones (about 20 percent). This commit
gives me back the lost estones.
|
|
Mark the obsoleted instructions bs_start_match2, bs_save2, bs_restore2, and
bs_context_to_binary as cold.
Remove support of a Y operand for bs_save2 and bs_restore2.
|
|
get_tuple_element with an Y register has become more frequent with
the new compiler.
|
|
The compiler used to generate "move Literal y(Y)" instructions very
rarely. Therefore, there was a transformation to avoid having a "move
c y" instruction.
With the new compiler, "move Literal y(Y)" instructions are relatively frequent, so we will need a "move c y" instruction.
|
|
|
|
|
|
|
|
* maint:
stdlib: Correct and optimize pretty printing of strings
|
|
* hasse/stdlib/optimize_pretty_printing/OTP-15639:
stdlib: Correct and optimize pretty printing of strings
|
|
* john/compiler/validator-type-conflict-fix/ERL-867:
beam_validator: Express test_arity/is_tagged_tuple as type tests
beam_validator: Fix type subtraction on select_* and inequality
beam_validator: Style fix for asserts in list comprehensions
beam_validator: Infer types from bs_put instructions
beam_validator: is_map no longer needs special treatment
beam_validator: Refactor type conflict resolution
beam_validator: Refactor branch handling
beam_validator: Fix element/2 BIF handling
beam_validator: Rename get_move_term_type and clean up get_raw_type
|
|
Also adds some more aliases that contain the key length
in their name.
|
|
OTP-15602
It is allowed to set file modes for the handler to use when opening
its log file. The given modes were earlier accepted without any
checks, which could make the handler behave unexpectedly. This commit
makes sure that
* if none of write, append or exclusive is given, then append is added
* if raw is not given, it is added
* if delayed_write or {delayed_write,_,_} is not given, then
delayed_write is added
|
|
|
|
|
|
This ensures that unreachable branches are properly ignored on
repeated checks (although tuple type subtraction isn't complete
yet).
|
|
Type subtraction never resulted in the 'none' type, even when it
was obvious that it should. Once that was fixed it became apparent
that inequality checks also fell into the same subtraction trap
that the type pass warned about in a comment.
This then led to another funny problem with select_val, consider
the following code:
{bif,'>=',{f,0},[{x,0},{integer,1}],{x,0}}.
{select_val,{x,0},{f,70},{list,[{atom,false},{f,69},
{atom,true},{f,68}]}}.
The validator knows that '>=' can only return a boolean, so once it
has subtracted 'false' and 'true' it killed the state because all
all valid branches had been taken, so validation would crash once
it tried to branch off the fail label.
|
|
|
|
|
|
|
|
into maint-21
* lukas/erts/fix_inet_buffer_auto_adjust/OTP-15651/OTP-15652:
kernel runtime dependency to erts
erts: Add yield via timeout to inet read_packet
erts: Don't increase buffer when sctp sndbuf is set
erts: Only change inet buffer if not set
|
|
* rickard/make-fixes-21/OTP-15551:
Fix install phase in build system
|
|
* hans/ssh/cuddle_tests:
crypto: Filter test case ciphers for supported ones
|
|
Change code to not send a message to itself as this allows other
messages to come in between. A change in TLS code made this scenario
more likely to happen, and the symptom is that the ftp:chunk_recv/1
function hangs.
|
|
* rickard/bootstrap-build-fix:
Fix build of bootstrap
|
|
|
|
|
|
|
|
|
|
The current type conflict resolution works well for the example
case in the comment, but doesn't handle branched code properly,
consider the following:
{label,2}.
{test,is_tagged_tuple,{f,ignored},[{x,0},3,{atom,r}]}.
{allocate_zero,2,1}.
{move,{x,0},{y,0}}.
%% {y,0} is known to be {r, _, _} now.
{get_tuple_element,{x,0},2,{x,0}}.
{'try',{y,1},{f,3}}.
%% ... snip ...
{jump,{f,5}}.
{label,3}.
{try_case,{y,1}}.
%% {x,0} is the error class (an atom), {x,1} is the error term.
{test,is_eq_exact,{f,ignored},[{x,0},{y,0}]}.
%% ... since tuples and atoms can't meet, the type of {y,0} is
%% now {atom,[]} because the current code assumes the type
%% we're updating with.
{move,{x,1},{x,0}}.
{jump,{f,5}}.
{label,5}.
%% ... joining tuple (block 2) and atom (block 3) means 'term',
%% so the get_tuple_element instruction fails to validate
%% despite this being unrechable from block 3.
{test_heap,3,1}.
{get_tuple_element,{y,0},1,{x,1}}.
{put_tuple2,{x,0},{list,[{x,1},{x,0}]}}.
{deallocate,2}.
return.
This commit kills the state on type conflicts, making unreachable
instructions truly unreachable.
|
|
While complex_test made certain branching instructions a lot easier
to read, we're still using `branch_state` for many others which is
hard to read and makes it impossible to "abort" branches on type
conflicts.
This commit replaces nearly all uses of `branch_state` with a
general branching mechanism, improving readability and paving the
way for proper type conflict resolution.
|
