| Commit message (Collapse) | Author | Age |
| |\ |
|
| | |
| |
| |
| | |
Plays better with apgdiff
|
| | |
| |
| |
| | |
Line insert is only a single operation with no new entities.
|
| | | |
|
| | |
| |
| |
| |
| | |
Entity value is MD5 hashed same as DB unique key, but the id itself is
now taken from the DB primary key which is sequence generated.
|
| | | |
|
| |/
|
|
| |
Or the last errno on failure.
|
| | |
|
| |
|
|
| |
Also logs them on main loop exit.
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
Adds virtual log function, real implementation writes to syslog.
Test implementation writes to BOOST_TEST_MESSAGE, perf implementation
discards.
Replaces existing prints to stderr and adds logs to all key points.
|
| |
|
|
|
|
|
|
|
|
| |
Store log lines in memory until threshold is reach or idle occurs, then
insert all the lines in a single transaction. Save points handle the
case of insertion errors. On success the queue is cleared.
Parked lines also saved in bulk, only necessary if queued lines could
not be inserted on shutdown, else the queue simply grows until ability
to insert is restored. Importing parked lines just adds them to the
queue and the normal process then follows.
|
| |
|
|
|
| |
Apache sends SIGTERM to the logger process to it shutdown. Honestly I
thought it would just close stdin and I should have checked.
|
| | |
|
| |
|
|
|
|
|
| |
UNIQUE CONSTRAINT is limited to 2704 bytes, which prevents inserting
large values. Here we swap to a unique index on the MD5 hash of the
value. This should more than suffice given we already map to a 32bit for
the id and the index size is much much smaller.
|
| |
|
|
| |
Replaces accidentally duplicated user_agent for correct content_type.
|
| |
|
|
|
| |
Easier checking if a job has completed [successfully] and reseting state
for the next time.
|
| |
|
|
|
| |
Neither the curl handle, not the operation map is thread safe. This
isn't ideal, but it does solve the problem in a safe manor.
|
| |
|
|
|
| |
Jobs run on background threads now, so we can happily run them even when
we're busy.
|
| | |
|
| | |
|
| |
|
|
|
|
| |
If that fails, we still park them as before, such as when the DB is
unavailable. Those which are saved as entities require investigation why
they couldn't be saved, much like UnparsableLines.
|
| |
|
|
| |
No changes.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
| |
Now a tuple of mapping functors and we pass each value through its
corresponding converter.
|
| |
|
|
|
| |
Replaces weird select with one thing with a function pointer stored in
the type definition array.
|
| | |
|
| |
|
|
|
| |
A pre-joined with entities view showing all the original data along with
ids; ideal for human readable stuff.
|
| | |
|
| |
|
|
|
|
|
| |
Don't persist entity ids saved to the DB until the transaction is
committed. Prevents the issue where a later DB operation fails, the
transaction is rolled back, but we still think the entity has been
saved.
|
| |
|
|
|
|
|
|
| |
Refactors CLFString in terms of QuotedString, but with the optional of
being null (nullopt)
Moves the whole decode function into QuotedString's parser, fixing
support for escaping of " which would otherwise prematurely end the
string in the middle.
|
| | |
|
| |
|
|
| |
Periodically, on idle, scan for and import previously parked lines.
|
| | |
|
| |
|
|
|
| |
oid is an "unsigned 4 byte integer", which matches our crc32 approach
perfectly, and is half the storage cost of bigint.
|
| |
|
|
|
| |
We call this parking, later we can reattempt ingestion after whatever
caused the failure has been fixed.
|
| |
|
|
|
|
|
| |
Holds all the settings and their defaults for use in program_options and
tests. Disables missing-field-initializers in tests because its over
sensitive to structures with defaults where you only provide some values
specifically.
|
| |
|
|
| |
Diagnostics and the ability to ingest later.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
Preparation step for having background curl operations.
|