- First:
Due to architectural constraints, it is very hard for the
input framework to handle optional records. For an optional record,
either the whole record has to be missing, or all non-optional elements
of the record have to be defined. This information is not available
to input readers after the records have been unrolled into the threading
types.
Behavior so far was to treat optional records like they are non-optional,
without warning. The patch changes this behavior to emit an error on stream-
creation (during type-checking) and refusing to open the file. I think this
is a better idea - the behavior so far was undocumented and unintuitive.
- Second:
For table and event streams, reader backend creation was done very early,
before actually checking if all arguments are valid. Initialization is moved
after the checks now - this makes a number of delete statements unnecessary.
Also - I suspect threads of failed input reader instances were not deleted
until shutdown
- Third:
Add a couple more consistency checks, e.g. checking if the destination value
of a table has the same type as we need. We did not check everything in all
instances, instead we just assigned the things without caring (which works,
but is not really desirable).
This change also exposed a few bugs in other testcases where table definitions
were wrong (did not respect $want_record)
- Fourth:
Improve error messages and write testcases for all error messages (I think).
If the input framework was used to read event streams and
those streams contained records with more than one field, not all
elements of the threading Values were cleaned up.
The reason for this is, that the SendEventStreamEvent function
returned the number of record elements in the outmost record
instead of the number of unrolled elements in the whole vector.
This number is later used to determine how many objects to delete.
Also - add a whole bunch of leak checks for the input framework
that would have caught that and should cover quite a number of
use-cases.
Replaced some with InternalWarning or InternalAnalyzerError, the later
being a new method which signals the analyzer to not process further
input. Some usages I just removed if they didn't make sense or clearly
couldn't happen. Also did some minor refactors of related code while
reviewing/exploring ways to get rid of InternalError usages.
Also, for TCP content file write failures there's a new event:
"contents_file_write_failure".
Closes#1021.
* origin/topic/bernhard/input-update:
this event handler fails the unused-event-handlers test because it is a bit of a special case.
...and fix the event ordering issue. Dispatch != QueueEvent
add Terminate to input framework to prevent potential shutdown race-conditions.
fix warning.
fix stderr test. ls behaves differently on errors on linux...
small fixes.
linux does not have strnstr
and close only fds that are currently open (the logging framework really did not like that :) )
A bunch of more changes for the raw reader
make reading from stdout and stderr simultaneously work.
allow sending data to stdin of child process
Streaming reads from external commands work without blocking anything.
replace popen with fork and exec.
change raw reader to use basic c io instead of fdstream encapsulation class.
* origin/topic/bernhard/thread-cleanup:
and just to be really sure - always make threads go through OnWaitForStop
hopefully finally fix last interesting race-condition
it is apparently getting a bit late for changes at important code...
spoke to soon (forgot to comment in line again).
Change thread shutdown again to also work with input framework.
Changing semantics of thread stop methods.
Support for cleaning up threads that have terminated.
Closes#997.
* topic/robin/sqlite-merge: (25 commits)
Fix to make sqlite test consistent, and updating coverage baselines
Avoid a CMake warning about 3rdparty looking like a number.
Fixing linker error.
and there is no has-reader.
make sqlite3 executable required and add test-cases for errors
Renaming src/external -> src/3rdparty
fix a few small rough edges (mostly comments that do no longer apply)
fix bug in input-manager regarding enums that a writer reads without 0-terminating the string
actually make sqlite work again (tests passed because the writer was not actually defined because of the define.)
add sqlite distribution.
fix warnings, update baselines, handle rotation
add sqlite tests and fix small vector/set escaping bugs
fix small bug with vectors and sets.
make work with newer AsciiFormatter.
start adding a different text for empty records for the sqlite writer.
no, you will never guess from where I copied this file...
make sqlite support more or less work for logging and input
make sqlite-writer more stable.
make it compile with new version of AsciiInputOutput
and adapt to AsciiInputOutput - seems to work...
...
Conflicts:
scripts/base/frameworks/input/__load__.bro
src/CMakeLists.txt
src/input.bif
src/input/Manager.cc
src/main.cc
src/types.bif
testing/btest/Baseline/coverage.bare-load-baseline/canonified_loaded_scripts.log
testing/btest/Baseline/coverage.default-load-baseline/canonified_loaded_scripts.log
Now it should work. However - this commit changes a basic assumption
of the threading queue. This basic assumption is, that nothing can
be read out of the out-queue of a dead thread. I think that reading
out of the queue of a dead thread makes perfect sense (when the thread
shuts down, pushes the rest of its work on the queue and says bye,
and wants the main thread to pick it up afterwards) - however, I
guess one can be of a differing opinion here.
In any case, it makes stuff a bit easier to understand - in my opinion.
It took me a while to find out why the messages disappear in thin
air and never arrive in the main thread ;)
Seems to work, tests pass, but not really verified.
Major change 1:
finished flag in MsgThread was replaced by 2 flags:
child_finished and main_finished.
child_finished is set by child_thread and means that the processing
loop is stopped immediately (no longer needed, no new input messages
will be processed, if loop continues running there is an ugly delay
on shutdown). (This took me a while to realize...)
main_finished is set by a message that is sent back by the child
to the main thread when Finished() is called (and child_finished
is set). when main_finished is set, processing of output messages
stops. But all messages that the child thread pushed in the queue
before calling Finish() are still processed.
Change 2:
Logging terminate call was replaced by a smaller call that just
flushes out the cache held by the main thread. This call
has to be done before thread shutdown is called - otherwhise
the threads will be shut down before all messages are pushed
on them. (This also took me a while to realize...).
Change 3:
Input framework actually calls it stop methods correctly (everything
was prepared, function call was missing)
* origin/topic/bernhard/sqlite:
fix a few small rough edges (mostly comments that do no longer apply)
fix bug in input-manager regarding enums that a writer reads without 0-terminating the string
actually make sqlite work again (tests passed because the writer was not actually defined because of the define.)
add sqlite distribution.
fix warnings, update baselines, handle rotation
add sqlite tests and fix small vector/set escaping bugs
fix small bug with vectors and sets.
make work with newer AsciiFormatter.
start adding a different text for empty records for the sqlite writer.
no, you will never guess from where I copied this file...
make sqlite support more or less work for logging and input
make sqlite-writer more stable.
make it compile with new version of AsciiInputOutput
and adapt to AsciiInputOutput - seems to work...
make it compile
add SQLite reader.
...adapt to new api...
now the writer supports tables and vectors.
basic sqlite writer seems to work.
* send end_of_data event for all kind of streams
* send process_finished event containing exit code of child process for executed programs
* move raw-tests to separate directory
* expose name of input stream to readers
* better handling of some error cases in raw reader
* new force_kill option for raw reader which SIGKILLs progesses on exit
The ordering of events how they arrive in the main loop is a bit peculiar at the moment.
The process_finished event arrives in scriptland before all of the other events, even though
it should be sent last. I have not yet fully figured that out.
- 'when' statements were problematic when used in a function/event/hook
that had local variables with an assigned function value. This was
because 'when' blocks operate on a clone of the frame and the cloning
process serializes locals and the serialization of functions had an
infinite cycle in it (ID -> BroFunc -> ID -> BroFunc ...). The ID
was only used for the function name and type information, so
refactoring Func and subclasses to depend on those two things instead
fixes the issue.
- 'return when' blocks, specifically, didn't work whenever execution
of the containing function's body does another function call before
reaching the 'return when' block, because of an assertion. This was
was due to logic in CallExpr::Eval always clearing the CallExpr
associated with the Frame after doing the call, instead of restoring
any previous CallExpr, which the code in Trigger::Eval expected to
have available.
- An assert could be reached when the condition of a 'when' statement
depended on checking the value of global state variables. The assert
in Trigger::QueueTrigger that checks that the Trigger isn't disabled
would get hit because Trigger::Eval/Timeout disable themselves after
running, but don't unregister themselves from the NotifierRegistry,
which keeps calling QueueTrigger for every state access of the global.
* origin/fastpath:
Fixing variable size issues (was uint, should be long) with http response code.
Revert "Trick for parallelizing input framework unit tests."
add opaque type-ignoring for the accept_unsupported_types input framework option.
* origin/fastpath:
...and forgotten debug-output, sorry (was already merged in some other internal repositories before I noticed)
and another bug in the input framework: config table does not work (is not transmitted to the readers) because the initialization was done the wrong way round.
more cases.
It will now not only fire after table-reads have been completed,
but also after the last event of a whole-file-read (or whole-db-read, etc.).
The interface also has been extended a bit to allow readers to
directly fire the event should they so choose. This allows the
event to be fired in direct table-setting/event-sending modes,
which was previously not possible.
without a final \0 - which means that strings read by the input framework are
unusable by basically all internal functions (like to_count).
the basic test now also checks this.
Thanks at Sheharbano for noticing this.
* origin/fastpath:
Ok, this one is not really necessary for 2.1 and more of a nice-to-have
another small bug found while searching for something else...
Fix two little bugs:
sorry. the patch for the set_separator.
make set_separators different from , work for input framework.
Bug found bei Keith & Seth: input framework was not handling counts and ints out of 32-bit-range correctly.
