* in general expand collection now performs only 1 sql
query to per expand field to fetch all items by the ids
instead of fetching them for every single collection item,
that should significantly increase performance in case
of large databases and reduce work for the database
* introduce $c->stash->{expand_cache} that contains
cached data for the expand fields to avoid multiple
same calculations
* expand_field() and expand_field_data() have been reworked
to support expand_cache
* new method expand_prepare_collection() is called for all
API GET collection methods before preparing resource hal
fields, to change the expand logic to only cache the data
instead of fetching it from the database
* new method expand_collection_fields() that is called in all
API GET collection methods after the @embedded data is prepared
to finalise the expand collection fields
* for expand collection there is only 1 SQL request per expand field
that fetches all items -in [ids to expand] and then the
expand_collection_fields() uses the cached items_by_ids with
O(1) fetch from the cache by id
Change-Id: Ie7c6115472878febf0d8c9b4d833f5c23b15c78b
* Role/Entities*: POST/PUT/PATCH/DELETE methods changes:
- support deadlock detection and transaction retry
(2 retry attempts at the moment)
- improve transaction control, use local $guard
instead of saving the ref to $c->stash, as in that
case it went out of scope too late and also reported
an error message into the log about abnormal $guard out
of scope interruption
- move all non transaction related code outside
of the scope
- add error handling when methods such as update_item,
and a like do not return the expected data, instead of
simply going out of scope and resulting in an uncontrolled
reply
* Role/API:
- rework transaction control:
+ get_transaction_control() is renamed to start_transaction()
to better reflect what it does
+ complete_transaction() is renamed to commit_transaction()
+ remove unused %params arg
+ pass $guard into commit_transaction() instead of having it
stored as $c->stash->{transaction_guard) that caused the $guard
ref to be destroyed much late than expected
(there was also a typo as transaction_quard,
which is not relevant anymore with the changes
+ add check_deadlock() that is invoked when an exception is caught
or an $c->errors contain an error, and if the error message
represents a transaction error, the transaction block is
re-invoked via "goto TX_START"
- rework error():
+ it now accepts args as following:
($self, $c, $code, $message, @errors)
# code -> returned as HTTP code in the reply
# message -> returned as HTTP message in the reply
# errors -> contain errors for internal logging, last element often contains a DBIx exception
+ populates all @errors into $c->error so they are available on
demend in the code via $c->error or $c->last_error
+ $c->log->error is not invoked now as the errors become printed
in log_response()
- log_response() now prints collected errors from $c->error correctly
as a separate log line, that is alike to the other api logs so that
those can be looked up by the request's tx_id, also all errors are
now printed only into api.log
* Adjust all $self->error() calls in catch($e) to include $e as the last
argument, as well as the duplicate $c->log->error is removed from
those ocassions
* Remove all $c->log->error() calls as they are replaced with either
$self->error() (that logs it correctly into api.log) or
$c->error('err') that also adds it correctly into api.log
* API::CallForwards: rework to use Entities/EntitiesItem
* API::Contracts: rework POST to use Entities
* API::PeeringGroups: rework POST to use Entities
* API::SubscriberRegistrations: rework POST to use Entities
* API::RewriteRuleSets: improve create_item() functionality
* Utils/Message: add 'api_retry' log type
* $c->session->{api_request_tx_id} is changed to
$c->stash->{api_request_tx_id} because sometimes the session
ref is different and a different tx_id becomes used
Change-Id: I633ce7a8047b1bf00a2f6889003088edf0825dcd
Kirill Solomko
Flaviu Mates