Improve performance of database::update_expired_feeds()

[abitmore]

According to profiling data mentioned in #1083, update_expired_feeds() plays a significant role while replaying.

---------------------- first 27 M blocks ----------------------------
764718ms th_a db_management.cpp:124 reindex ] Done reindexing, elapsed time: 5063.73962899999969522 sec

Flat profile:

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total
 time   seconds   seconds    calls   s/call   s/call  name
  9.62    215.94   215.94 448458421     0.00     0.00  graphene::chain::generic_index<graphene::chain::account_object, ... >::find(graphene::db::object_id_type) const
  8.47    406.11   190.17 20731743387     0.00     0.00  graphene::chain::operator>(graphene::chain::price const&, graphene::chain::price const&)
  7.18    567.18   161.07 390785433     0.00     0.00  graphene::chain::database::adjust_balance(graphene::db::object_id<(unsigned char)1, (unsigned char)2, graphene::chain::account_object>, graphene::chain::asset)
  4.68    672.30   105.12 307448392     0.00     0.00  graphene::chain::generic_index<graphene::chain::account_statistics_object, ... >::find(graphene::db::object_id_type) const
  4.64    776.50   104.20 22967573183     0.00     0.00  graphene::chain::operator<(graphene::chain::price const&, graphene::chain::price const&)
  4.39    874.95    98.45                             sha256_block_data_order_avx
  3.92    962.82    87.87 3183637020     0.00     0.00  graphene::chain::generic_index<graphene::chain::asset_bitasset_data_object, ... >::find(graphene::db::object_id_type) const
  3.62   1044.04    81.22 386682199     0.00     0.00  graphene::chain::generic_index<graphene::chain::account_balance_object, ... >::modify(graphene::db::object const&, std::function<void (graphene::db::object&)> const&)
  3.32   1118.50    74.46 805850110     0.00     0.00  graphene::chain::generic_index<graphene::chain::asset_object, ... >::find(graphene::db::object_id_type) const
  2.85   1182.43    63.93 245765371     0.00     0.00  graphene::chain::generic_index<graphene::chain::account_statistics_object, ... >::modify(graphene::db::object const&, std::function<void (graphene::db::object&)> const&)
  2.76   1244.32    61.89 27000000     0.00     0.00  graphene::chain::database::update_expired_feeds()
  2.16   1292.90    48.58 72953692     0.00     0.00  graphene::chain::generic_index<graphene::chain::limit_order_object, ... >::create(std::function<void (graphene::db::object&)> const&)

• The 11th entry database::update_expired_feeds() is it itself;

• the 7th entry index<asset_bitasset_data_object>::find() is caused by it here:

  bitshares-core/libraries/chain/db_update.cpp

  Line 483 in 06aee78

  const asset_bitasset_data_object& b = a.bitasset_data(*this);

• the 9th entry index<asset_object>::find() is caused by it due to an unnecessary call here:

  bitshares-core/libraries/chain/db_update.cpp

  Line 494 in 06aee78

  check_call_orders(b.current_feed.settlement_price.base.asset_id(*this));

These 3 entries sum up to 10% of replay time.

Current code iterates through all asset_objects who is a bit asset or a prediction market on every new block, then fetch their asset_bitasset_data_object, then check if feed is expired. However, we have more than 500 bit assets / prediction markets on the chain now, it's inefficient to iterate through them all on every block.

Another issue is related to cryptonomex/graphene#615. For the first 5,000,000 blocks or so, update_median_feeds() is almost called for every bit asset on every block. Although the bug has been fixed with a hard fork, the buggy code is still there for processing blocks before the hard fork, thus wasting our time for every sync/replay.

To optimize, things need to be done:

• update median feed and check call orders when feed expired
  • add a by_expiration index in asset_bitasset_index, only iterate through and process expired ones
  • refactor pre-HF615 code with better performance
• update CER in asset_object when
  • median CER changed, or
  • CER in asset_object got updated (E.G. by asset_update_operation)

This is a sub-task of #982.

[abitmore]

Almost done on top of #1099.

Total replay time for 27,000,000 blocks reduced from 5063 seconds to 3761 seconds, the difference is 1 - 3761/5063 ~= 25%.

Will create pull request after done cleanup.

[oxarbitrage]

excellent and very appreciated!

[abitmore]

Done with #1180.