On MySQLdb (1.2.2) we still experience stale uncommitted transactions and different warden_server processes don't see actual db data. Looks like there is at least one stale transaction per daemon, appearing at any call to execute. I have no better idea than to implicitly commit after each SELECT (INSERTs are committed/rolled back explicitly).

* filenames now have .idea extension (Mentat requires that :) )
* bugfix: use list range instead of generator xrange on event index list generation (also in warden_client)

Brown paper bag commit

This reverts commit 1fc362c3.

* warden_server: Added "events_id" key into errors along with "events" - makes error messages bigger, but allows client operators to identify offending messages by stable identifiers based on logs.
* Removed too much internal info from database errors.
* Removed format type checking from Draft4Validator, using only explicit schema regexps - Draft4Validator raised FormatError istead of ValidationError, rendering iter_errors unusable.
* warden_client: Implemented retries on server errors.
* Client now honours send_events_limit from server - too long list of events is split and sent in chunks.
* getInfo and sendEvents update local send_events_limit according to one sent by server in info or error message.
* Logging is now explicit, not authomatic in Error class, allowing to log only single time at top level methods.
* More sane default client name.
* Errors from server are now checked and ensure correct format.
* warden_filer: Ditched error and retry handling, warden_client now does for us.

Reworked error handling in server, client and filer (again :( ). We definitely have to support multiple errors so let's do it right.  Error now contains list of errors.  Got rid of "detail" section everything now goes directly within particular error.  Error can bear "events" attribute for list of sent events concerned (or none if all sent events erred). Streamlined error logging.
Server now retries if MySQL went away.
Response result now gets properly encoded if Unicode appears somehow (however that's just fighting symptoms, should get more thorough fix).

plain files in directory

syntax highlighting. :) (deja-vu)

or rolled back. insertLastReceivedId was not committed, and as we reuse the cursor, reader
client kept getting stale data until next event insert on the same connection, possibly
from completely different client at completely unrelated time (which unintentionally
closed even the previous transaction), or until server restart. Whew, tricky beast to hunt.

syntax highlighting. :)

  consistently. Possible keeping and reopening where possible will need more
  work and testing, I'm not keen to try to clench in now. So - changed to
  reopening connection each time, which will possibly affect performance (on
  small queries).
- Both server and client library now defaults to get_events_limit instead of 1

Fixed Recon id in catmap
Fixed isoformat calling on None
Shortened validation error message
Fixed rejecting of Test messages

Parent categories are now also saved in event_category_mapping, which simplifies fetch_events. Fixed AND->OR in group checking. Group now also matches only at "." boundary.

Added optional client auth by shared secret (also modified schema, client, curl client). Modified Client __str__ method for log display without disclosing secret. Modified wsgi server logging and parameter disposing accordingly.

Clients are now authorized by client= argument in URL, schema and client lib is updated accordingly. Method authentication defined in "expose" decorator. Added Client container class for deterministic logging. Schema update to consistent InnoDB and utf8mb4 everywhere.

Auth for sending events fixed

Error is now raised via req.error helper, which automatically adds already known info (method, request id), thus calls are terser.
Client's Error communicates request id, if received from server.
exc_info made explicit everywhere. Automatic exc_info gathering sometimes returned bogus info about unrelated exception from far before.

Logging now automatically adds this request id + called method into all log messages.

Added Request class for request data shared among instances. Will simplify error handling and logging

Server is now less revealing in error messages to clients (local logging is still as verbose as requested)

Fixed multiple HTTP args encoding/parsing, store_event now uses SQL placeholders, more robust fetching of cat/tag in store_event, simplified map_id usage

Now auth/auth actually works. Now uses services on many places, where clients was used (and didn't work)

Fixed for now: /opt/warden_server_3/warden_server.py:427: Warning: Field 'request' doesn't have a default value