Use Curve Prices API for pool volume

[nagakingg]

Description

Updates historical pool volume functionality to use the new Curve Prices API. Pool volume for each token pair is denominated in the pair's quote token, rather than pseudo-dollars, to allow for correct volume limiting for v2 pools. In the volume-limited arbitrage pipeline, volume multipliers are computed for each pair using this data. Volume limiter modes are removed as this method is more accurate than all of the previous modes, which relied on aggregate pool volume across all pairs.

• Add network.curve_prices
• Add ApiResultError to exceptions
• Add vol_limited_arb.pool_volume.get_pool_volume()
• Compute volume multiplier per asset pair
• Remove volume multiplier modes
• Remove pipelines.utils
• Remove PriceVolume.total_volumes()
• Remove volume from PoolDataCache

Hygiene checklist

• Changelog entry
• Everything public has a Numpy-style docstring
  (modules, public functions, classes, and public methods)
• Commit history is cleaned-up with minor changes squashed together
  and descriptive commit messages following Tim Pope's style

Cute Animal Picture

[nagakingg]

A few open questions:

• Should PoolDataCache be fully removed in this PR? Or hold off
• Should vol_limited_arb.pool_volume.get_pool_volume() be moved to pool_data.get_pool_volume()? If so, I think it should also allow a pool address to be input in lieu of a pool metadata object

Edit:

• PoolDataCache fully removed
• get_pool_volume moved to pool_data

[github-actions]

Coverage report

The coverage rate went from 76.03% to 78.08% ⬆️
The branch rate is 64%.

91.15% of new lines are covered.

Diff Coverage details (click to unfold)

curvesim/exceptions/init.py

100% of new lines are covered (100% of the complete file).

curvesim/network/curve_prices.py

91.42% of new lines are covered (86.84% of the complete file).
Missing lines: 47, 48, 122

curvesim/pipelines/vol_limited_arb/init.py

16.66% of new lines are covered (36.36% of the complete file).
Missing lines: 94, 103, 104, 105, 106

curvesim/pool/init.py

0% of new lines are covered (72.63% of the complete file).
Missing lines: 279

curvesim/pool_data/init.py

100% of new lines are covered (100% of the complete file).

curvesim/pool_data/queries/init.py

100% of new lines are covered (100% of the complete file).

curvesim/pool_data/queries/metadata.py

100% of new lines are covered (88% of the complete file).

curvesim/pool_data/queries/pool_volume.py

98.24% of new lines are covered (95.45% of the complete file).
Missing lines: 93

[nagakingg]

Comparing 3pool sim results before and after the volume limit change. As expected:
(a) Old version volume limits are roughly the average of the new version volume limits
(b) the results are similar but not exactly the same

Volume Limits
Old version:
DAI/USDC: 0.000037
DAI/USDT: 0.000037
USDC/USDT: 0.000037

New version:
DAI/USDC: 0.000016
DAI/USDT: 0.000022
USDC/USDT: 0.000054

Results
Old Version:

New Version:

[chanhosuh]

Everything looks really nice, big improvements on the current codebase!

I would update the changelog entries tho :)

[chanhosuh]

We should mention we've removed volume support for chains other Mainnet (pending future support from the curve-prices API) and sims for the RAI pool are no longer supported.

[chanhosuh]

I understand stuff gets aggregated for the release changelog, but it helps to leave out clutter like this: nobody will ever care you added this error class, but probably most people would want to know what network.curve_prices is.

[chanhosuh]

I'm not a fan of imports like this because they couple tests together unexpectedly. Nobody working on pool metadata tests would expect the pool volume integration tests to fail because they changed something.

Probably we should have a better test setup of some sort, whether it's fixtures or something else, but from what we discussed, this will go away anyway due to volume functionality being injected into a ReferenceMarket class, so I don't think it's worth it.

In that case, I would expect better intermediate abstractions of some sort, e.g. pool metadata --> asset abstractions --> volume functions; this would be "happening" inside a ReferenceMarket (or its factory), but for testing, it would separate pool metadata rather cleanly from the volume functions via the intermediate asset abstractions.