| Age | Commit message (Collapse) | Author |
|---|
|
* hasse/remove_old_IO_requests/OTP-15695:
stdlib: Remove old I/O-requests from test suites
stdlib (doc): Do not mention old requests
snmp: Update old format I/O-request
sasl: Use encoding when reading terms
parsetools (doc): Update old format I/O-requests
mnesia: Update old format I/O-request
kernel: Update old format I/O-requests
stdlib: Remove no longer used function
kernel: Remove commented out code
stdlib: Remove commented out code
|
|
|
|
* dgud/mnesia/sticky-bug/ERL-768/OTP-15979:
mnesia: Bump protocol version
mnesia: Introduce sync_asym_trans protocol
|
|
Remove unsupported conversions from the old protocols.
This means mnesia on OTP-R18 nodes is no longer able to connect to
OTP-22 mnesia nodes.
Add protocol conversion for old nodes and sync_asym_trans transactions.
|
|
Transactions with sticky locks could with async_asym transactions be
committed in the wrong order, since asym transaction are spawned on
the remote nodes.
See ERL-768.
|
|
If add_table_copy was called when a node was starting it deadlock
waiting for mnesia_controller, when schema was not merged.
Abort if that is the case.
|
|
If an activity function fails with an error, the stacktrace is
dropped in non_transaction function try/catch clause.
This makes debuging of errors inside a transaction really hard.
Transaction activities handle errors and exits differently, non-transaction
activities should do the same.
Adding the stacktrace to the exit reason when translating errors to exits.
|
|
- first stab at fixing index_[read|match_object] w plugins
- do not include index plugins at match ops
- add index_plugin_test_suite
- Increase ix plugin test coverage + bug fix
|
|
* maint:
mnesia: Avoid raise in overload reports
|
|
Several processes could invoke adjust_log_writes, which
could cause overload reports.
|
|
Optimise ext table dumping
OTP-15588
|
|
The original algorithm first grouped ops in a commit by ext engine
type via folding over the list with orddict:append/3, that resulted in
an O(n^2) algorithm.
However, grouping the ops is not needed, the ops can be dispatched to
insert_op/6 one-by-one, after looking up the storage semantics of
their respective engine. This is a much cheaper algorithm, assuming
looking up the storage semantics is cheap (which should be).
|
|
|
|
Do not read from (set | order_set) source table if the content is
already in the transactional context (i.e., a previous write/delete).
|
|
* maint:
Updated OTP version
Prepare release
Updated the engine load functionality
Clean up bag index tables
|
|
* maint-20:
Updated OTP version
Prepare release
Updated the engine load functionality
Clean up bag index tables
Conflicts:
OTP_VERSION
lib/crypto/c_src/crypto.c
lib/crypto/doc/src/notes.xml
lib/crypto/vsn.mk
lib/mnesia/doc/src/notes.xml
lib/mnesia/vsn.mk
otp_versions.table
|
|
Index records for bag tables with ram_copies was not deleted
after "real" objects where deleted and thus a memory leak.
|
|
* maint:
Fix delete_object and write convergence in transaction.
Replace for loop with strchr
Fix typos
Clean unused wxe_driver callbacks
Remove wx c_src compilation warning on Darwin
|
|
* upstream/pr/1858:
Fix delete_object and write convergence in transaction.
OTP-15231
|
|
* maint:
Updated OTP version
Update release notes
Update version numbers
crypto: Fix crash in compute_key(ecdh, ...) on badarg
Relax add_table_copy restriction
Fixed #Ref ordering bug
Test #Ref ordering in lists and ets
Do NOT disc_load from ram_copies when master_node is set
ssl: Make sure that a correct cipher suite is selected
ssl: Correct handling of empty server SNI extension
|
|
* maint-20:
Updated OTP version
Update release notes
Update version numbers
crypto: Fix crash in compute_key(ecdh, ...) on badarg
Relax add_table_copy restriction
Fixed #Ref ordering bug
Test #Ref ordering in lists and ets
Do NOT disc_load from ram_copies when master_node is set
ssl: Make sure that a correct cipher suite is selected
ssl: Correct handling of empty server SNI extension
|
|
* dgud/mnesia/add_table_copy_ram/OTP-15226:
Relax add_table_copy restriction
|
|
Allow to add replicas even if all other replicas are down when the
other replicase are not stored on disk.
|
|
Setting master_nodes to a node with ram_copies replica and
that node had not loaded the table, could cause it load an
empty table, even though (non master) nodes had disc_replicas.
This meant that tables where unexpected empty after multiple failures
happened. When this happen do not load the table and wait for user
to force_load it on some node, preferably with a disk copy.
|
|
I did not find any legitimate use of "can not", however skipped
changing e.g RFCs archived in the source tree.
|
|
|
|
|
|
Fix a bug, when delete_object was deleting the record if it was written
in the same transaction even if it was written to a different value.
To verify:
%% Create a set table
mnesia:create_table(foo, []).
%% Write and delete_object in transaction
mnesia:transaction(fun() ->
mnesia:write({foo, bar, one}),
mnesia:delete_object({foo, bar, not_one})
end).
{atomic, [{foo, bar, one}]} = mnesia:transaction(fun() -> mnesia:read(foo, bar) end).
Added a section to isolation tests to check for non-matching delete_object
requests.
|
|
|
|
make erlang:process_info/1 not retrieve messages
|
|
process_info/1 retrieves a number of properties related to a process,
including the list of messages in its mailbox. This is potentially
unsafe if the target process has a large number of queued messages:
- there is no a priori upper bound on the amount of memory being
allocated to hold that list, and
- the loop to retrieve the messages is uninterruptible, so the
Erlang scheduler where this executes blocks for the duration
We've seen process_info/1 bring down heavily loaded nodes on more
than one occasion. At least once it appeared to have blocked the
Erlang heart process from executing, causing the external heart to
kill the VM.
Consequently this removes 'messages' from the list of process_info
tags to retrieve for process_info/1. Note that process_info/1 still
retrieves 'message_queue_len', and process_info/2 can still retrieve
'messages' when asked to.
A few places in the OTP libraries need minor adjustments, since they
want 'message_queue_len' but compute it from the length of the list
of messages.
|
|
Take care of the call of erlang:get_stacktrace() in
module mnesia_lib.
|
|
The remaining call (in module mnesia_lib) is taken care
of in the following commit.
|
|
* maint:
Updated OTP version
Prepare release
mnesia: Fix checkpoint crash
|
|
* dgud/mnesia/checkpoint-crash/OTP-14841:
mnesia: Fix checkpoint crash
|
|
* dgud/mnesia/slow-startup/OTP-14829:
mnesia: Read schema user properties directly
|
|
Bad timing can cause retain messages to go to a new process if
checkpoint name is reused directly and the checkpoints contain
different tables.
Ignore those messages instead of crash.
|
|
* maint:
Updated OTP version
Prepare release
mnesia: Read schema user properties directly
ssh: testcases for space trailing Hello msg
ssh: Don't remove trailing WS in Hello msg
|
|
* maint-19:
Updated OTP version
Prepare release
mnesia: Read schema user properties directly
ssh: testcases for space trailing Hello msg
ssh: Don't remove trailing WS in Hello msg
|
|
* dgud/mnesia/slow-startup/OTP-14829:
mnesia: Read schema user properties directly
|
|
Doing a proper transaction handling of user properties (in current
implmenentation) caused schema_transactions to be quadratic which
caused huge startup times for nodes with many tables since the merge schema
is a schema_transaction.
Do a direct read instead, cheap but implies that the plugin changes
will not be visible until the changes are commited and cannot be used
later in same transaction, which was possible before if undocumented
functions where used.
|
|
* maint:
mnesia: Fix error handling in abort write
|
|
* dgud/mnesia/backup-error-handling/OTP-14776:
mnesia: Fix error handling in abort write
|
|
* maint:
Fix type create_option() in mnesia
|
|
The create_option() type in mnesia lacks the user_properties field
which is used by mnesia in various places.
|
|
|
|
The "root cause" error reason was lost when abort_write failed,
e.g. file was already closed, and Mod:abort_write(..) returned an
error.
|
|
~tw and new string functions are new since OTP-20 (stdlib-3.4)
|
|
|
|
|
