HIP 109 boost by device type #798

Making the internal member hexes private will make it easier to change the implementation when device type is introduced.

A cell can be boosted more than once. Old boosts will retain device_type of ::All. Any boost matching the device_type accumulates into the multiplier. ex: 3 boost with a multiplier of 10, will multiply by 30.

The device_type column is assumed to be a jsonb, because the device_type column for the `mobile_hotspot_infos` table is an enum, and a jsonb field. It's nullable, which will represent boosting a hex for ALL device types. Otherwise, a camelcase value mapping to the DeviceTypeV0 in helium-program-library. https://github.com/helium/helium-program-library/blob/master/programs/hexboosting/src/state.rs This is different from the snake_case values used with protobufs.

With type hints enabled, `hex` shows up as a `BoostedHex` going into a function that expects `BoostedHex`. Renaming to `hex_proto` will hopefully clear up the need for the type casting.

This test uses a single type of radio to reduce adding noise to rewards by factors other than hex boosting. The first pass of the test calculates rewards for 2 radios with no boosted hexes to prove they are in equal standing. The second pass boosts only the hex for single radio, but with multiple BoostedHexDeviceType's to ensure they accumulate.

There was not enough custom functionality to warrant owning a BoostedHexDeviceType enum of our own.

* Refactor to use BoostedHexes methods Making the internal member hexes private will make it easier to change the implementation when device type is introduced. * Add device type to boosted hex info * refactor metadata_db tests to make test clearer also makes it easier to add new tests * remove expired boosted hexes when streaming from db * ensure no tests are written with expired boosted hexes * optimize by computing end_ts in db query Thanks for the query help Brian! By precomputing the end timestamp of a boosted hex, we can not have to stream all the hexes out of the db just to throw them away. * fixup after rebase - remove unused imports - remove old refactor return types * make boosted hex test function more explicit If the expired check had been made a global check, the ability to use BoostedHexes for modified hexes would have broken at runtime. The attempt here is to make very explicit during testing how to meet the same contract as the database queries for boosted hexes. I think there are still some cracks, but we can narrow in on those as we find them. For now, I think naming test constructor functions is a good start.

There is another test for same device types across hexes. This test is for different types in the same hex. The 2nd unboosted hex serves the purpose of giving us a relative baseline so we don't need to hardcode values into the test.

It becomes more difficult to compare ratios with radio_reward_v2 because boosted rewards are scaled. In many of the test we were hoping to have nice ratios (1:10) of poc_rewards. However, this was mostly because a radio would have received 0 boosted_rewards. Checking that ratio is impossible. We can however check the coverage_points (now that they mean points, and not rewards) for the expected ratio, and rely on other tests to ensure we're calculating poc_reward amount correctly.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

HIP 109 boost by device type #798

HIP 109 boost by device type #798

Commits on Sep 6, 2024

HIP 109 boost by device type #798

Are you sure you want to change the base?

HIP 109 boost by device type #798

Commits on Sep 6, 2024