Protorev smarter logic and more testing #4181

NotJeremyLiu · 2023-01-31T15:32:18Z

What is the purpose of the change

This PR adds more testing to x/protorev and improves the arb calculation logic to be more efficient.

Brief Changelog

Given a route, we now calculate if the minAmountIn is profitable as a pre-check. If minAmountIn is not profitable, then the route is not profitable and we don't run the full binary search algo. This will make Protorev take much less compute time in production compared to its previous design (which would run the full binary search even when a route has no-arb). We only increment the pool point counter if the trade is profitable, thereby saving us compute for swaps that create no backrunning opportunity.
Given a route, we now calculate if the maxAmountIn and maxAmountIn + 1 profit is increasing as a pre-check and increase the range if so. If the profit from maxInputIn is still increasing, this means that we need to increase the range to capture the max profit.
Adds tests that verifies the limit/point system works properly, >3 pool routes work, and other denom besides osmo/atom work.
Deprecate hot routes from being included in genesis state for simplicity.
Add E2E testing that
- Checks that the module is correctly init on start
- Checks that the module can correctly spot a back running opportunity after a swap
- Checks that the module executes a profitable cyclic arbitrage trade
- Checks that the module did not mint or alter the supplies from the bank
Add support for gamm swap msgs
Deprecate atom from the list of base denoms that are used on genesis for route creation

Testing and Verifying

All previous tests pass with new changes
Added a test to ensure range extension works properly
Added a test for swapExactAmountOut
Added benchmark testing
Added doomsday testing to ensure point system works properly and doesn't allow protorev to keep running
Adds tests for generalized denoms
Adds a test to ensure panic catching is handled properly. This currently FAILS. This is on purpose, this PR is not to be merged until Panic Handling In MultihopEstimateOutGivenExactAmountIn borks Protorev #4180 is resolved
Add test for gamm msg swaps
Add tests for E2E (as described above)

NotJeremyLiu

Reviewed, left 1 comment about the changes made. Going to fix up the doomsday test given the new point refund logic.

NotJeremyLiu · 2023-02-14T20:34:59Z

x/protorev/keeper/rebalance.go

+	if err != nil {
+		return sdk.Coin{}, sdk.ZeroInt(), err
+	} else if minInProfit.LTE(sdk.ZeroInt()) {
+		return sdk.Coin{}, sdk.ZeroInt(), fmt.Errorf("no profit for route %d", route.Route.PoolIds())


I imagine a majority of small swaps will result in no profits, so this will throw an error often. Feels kinda like we want to view no profit in a route as a non-error. maybe we log in some other way if we want to log?

Resolved in ad5ea73

NotJeremyLiu · 2023-02-14T20:53:15Z

Another thing I noticed is that we still have a hardcoded denom for atom:

osmosis/x/protorev/types/constants.go

Line 8 in e6c1ba6

    
           var AtomDenomination string = "ibc/27394FB092D2ECCD56123C74F36E4C1F926001CEADA9CA97EA622B25F41E5EB2"

This is no longer used in protorev itself, but used throughout our test suite. Is there a way for us to move this easily to only be for test suite and not in any of the core module files (to avoid changing all the tests that reference it, but not have any unnecessary code in core module files).

I think this is a non-issue though

davidterpay · 2023-02-15T18:29:29Z

Another thing I noticed is that we still have a hardcoded denom for atom:

osmosis/x/protorev/types/constants.go

Line 8 in e6c1ba6

var AtomDenomination string = "ibc/27394FB092D2ECCD56123C74F36E4C1F926001CEADA9CA97EA622B25F41E5EB2"

This is no longer used in protorev itself, but used throughout our test suite. Is there a way for us to move this easily to only be for test suite and not in any of the core module files (to avoid changing all the tests that reference it, but not have any unnecessary code in core module files).

I think this is a non-issue though

Update, I deprecated it from the module.

davidterpay · 2023-02-15T18:34:01Z

proto/osmosis/protorev/v1beta1/genesis.proto

-  // Hot routes that are configured on genesis
-  repeated TokenPairArbRoutes token_pairs = 2 [ (gogoproto.nullable) = false ];


for reviewers: deprecating the use of hot routes in genesis since we will not be using it.

Hrm, I feel like it should be added back (in a subsequent non-release affecting PR), so that ExportGenesis -> ImportGenesis still works. (So state export based testnets for instance can use this)

The ideal is that for state' = import(export(state)), that state' is effectively the same as state.

davidterpay · 2023-02-15T18:38:04Z

tests/e2e/initialization/config.go

 	MinGasPrice         = "0.000"
 	IbcSendAmount       = 3300000000
 	ValidatorWalletName = "val"
 	// chainA
 	ChainAID      = "osmo-test-a"
-	OsmoBalanceA  = 200000000000
+	OsmoBalanceA  = 20000000000000


for reviewers: osmo balance being increased so that the pools that protorev testing creates have enough funds

davidterpay · 2023-02-15T18:47:29Z

tests/e2e/initialization/config.go

+	UstBalanceA   = 500000000000000
+	LuncBalanceA  = 500000000000000


for reviewers: denoms used in protorev testing

p0mvn · 2023-02-16T02:56:04Z

tests/e2e/initialization/config.go

@@ -52,24 +52,29 @@ const (
 	AtomDenom           = "uatom"
 	OsmoIBCDenom        = "ibc/ED07A3391A112B175915CD8FAF43A2DA8E4790EDE12566649D0C2F97716B8518"
 	StakeIBCDenom       = "ibc/C053D637CCA2A2BA030E2C5EE1B28A16F71CCB0E45E8BE52766DC1B241B7787"
+	UstIBCDenom         = "ibc/BE1BB42D4BE3C30D50B68D7C41DB4DFCE9678E8EF8C539F6E6A9345048894FCC"


Hi @davidterpay . Please manually backport just this file to v14.x branch.

Once merged, please ping me, and I will give you an image tag that will need to be put here:

osmosis/tests/e2e/containers/config.go

Line 30 in dd1d910

previousVersionInitTag = "v14.x-f6813bcb-1674140667-manual"

We do this to be able to build genesis and config files from v14.x to start up Osmosis containers of that version and perform the upgrade to the next (v15) version in your branch.

ValarDragon · 2023-02-16T16:15:32Z

x/protorev/types/constants.go

-var StepSize = sdk.NewInt(1_000_000)
+// ExtendedMaxInputAmount is the upper bound index for finding the optimal in amount
+// when determining route profitability for an arb that's above the default range (2 ^ 17) = 131,072
+var ExtendedMaxInputAmount = sdk.NewInt(131_072)

 // Max iterations for binary search (log2(16_384) = 14)


Suggested change

// Max iterations for binary search (log2(16_384) = 14)

// Max iterations for binary search (log2(131_072) = 17)

ValarDragon · 2023-02-16T16:16:47Z

x/protorev/keeper/epoch_hook_test.go

+
+	for i := 0; i < b.N; i++ {
+		b.StartTimer()
+		suite.App.ProtoRevKeeper.UpdatePools(suite.Ctx)


This needs enough pools setup to be meaningful, right?

Atm, the test suite sets up ~50 pools. I attached the benchmark results to the comment. if this scales linearly and we assume mainnet has around ~1000 pools, the execution time of UpdatePools should be around 5 ms in total.

Ah awesome! Didn't realize this was the same setup test with everything else, that makes sense!

Added: - 4 pool mainnet route test - 2 pool mainnet route test - Non-osmo/atom denom test

- Currently panics due to solveCFMMBinarySearchMulti in stableswap calculation

- Tests that the per tx limit works properly - Tests that the per block limit works properly

- Adds a pre-check before the binary search that tells us if we need to run the binary search at all - Reduces overall computation/time by avoiding binary searching profit amounts over routes without any profit opportunity

- Added "test/3" denom to test non-osmo/atom denoms in previous PR - didn't notice it broke one test that expected us to only have 2 denoms, so adding "test/3" here so everything passes

- This implements a way for us to have a binary search bound that can increase in size to account for large trades above the default range - I believe this is the minimum-amount-of-new-code approach, but may not be the optimal approach (what this method avoids is having to save the amount in or amount out of the original swap, and then backtracking/converting that amount to the base denom for an arb, and creating bounds off of that)

- Doubling isn't the wanted behavior (only wanted if lower bound is 1), just increasing bound by the same range size is wanted (so adding MaxInputAmount achieves this)

- Increases max iterations to 17 to allow for situation when we need to increase the upper bound - Add new ExtendedMaxInputAmount variable for us to use as this new max bound range - Replace bound changing logic from iteratively changing the range to immediately giving our max range

- Tests binary search range extension logic works properly

- This currently fails, this is on purpose so we don't merge the PR and think we are good until this passes and the panics are handled properly

- Makes pool 41 reserves to have an arb opportunity in the doomsday routes - Changes tx limit and block limit specifically for doomsday testing

ValarDragon · 2023-02-16T16:24:42Z

x/protorev/genesis.go

 		panic(err)
 	}

-	// Update the pools on genesis
-	if err := k.UpdatePools(ctx); err != nil {
+	// Setting the admin account by default to a trusted address


Can we document the source of this address? (Perhaps link to the commonwealth post once it exists)

(Lets make a followup issue for this)

Yup, will do. Just for reference this is an account we created and are storing on a ledger. We will use it to enable authz (which we have already test) for a different account so that the admin keys are safely stored.

tests/e2e/configurer/chain/queries.go

ValarDragon · 2023-02-16T18:00:40Z

x/protorev/keeper/keeper_test.go

+		{ // Pool 41 - Used for doomsday testing
+			initialLiquidity: sdk.NewCoins(
+				sdk.NewCoin("usdc", sdk.NewInt(1000000000000000)),
+				sdk.NewCoin("usdt", sdk.NewInt(1000000000000000)),
+				sdk.NewCoin("busd", sdk.NewInt(2000000000000000)),
+			),
+			poolParams: stableswap.PoolParams{
+				SwapFee: sdk.NewDecWithPrec(1, 4),
+				ExitFee: sdk.NewDecWithPrec(0, 2),
+			},
+			scalingFactors: []uint64{1, 1, 1},
+		},


Thanks for adding all these pools to whats protorev configured in the tests!

ValarDragon · 2023-02-16T18:06:15Z

x/protorev/keeper/rebalance.go

-		inputCoin, profit, err := k.FindMaxProfitForRoute(ctx, route, inputDenom)
-		if err != nil {
-			k.Logger(ctx).Error("Error finding max profit for route: ", err)
+	// Get the total number of pool points that can be consumed in this transaction


Nice job on all the code comments in this file, quite easy to follow the logic :)

ValarDragon · 2023-02-16T18:06:58Z

x/protorev/keeper/rebalance.go

+	// If a cyclic arb exists with an optimal amount in above our minimum amount in,
+	// then inputting the minimum amount in will result in a profit. So we check for that first.
+	// If there is no profit, then we can return early and not run the binary search.
+	_, minInProfit, err := k.EstimateMultihopProfit(ctx, inputDenom, curLeft.Mul(route.StepSize), route.Route)


ValarDragon

Nice job on this! LGTM to merge!

* Add generalized tests Added: - 4 pool mainnet route test - 2 pool mainnet route test - Non-osmo/atom denom test * Add stableswap doomsday test - Currently panics due to solveCFMMBinarySearchMulti in stableswap calculation * Add tests for pool point limits - Tests that the per tx limit works properly - Tests that the per block limit works properly * Add pre-check before binary search - Adds a pre-check before the binary search that tells us if we need to run the binary search at all - Reduces overall computation/time by avoiding binary searching profit amounts over routes without any profit opportunity * Add denom to make test pass - Added "test/3" denom to test non-osmo/atom denoms in previous PR - didn't notice it broke one test that expected us to only have 2 denoms, so adding "test/3" here so everything passes * Implement minimum change to have smarter binary search bounds - This implements a way for us to have a binary search bound that can increase in size to account for large trades above the default range - I believe this is the minimum-amount-of-new-code approach, but may not be the optimal approach (what this method avoids is having to save the amount in or amount out of the original swap, and then backtracking/converting that amount to the base denom for an arb, and creating bounds off of that) * Switch range increasing logic - Doubling isn't the wanted behavior (only wanted if lower bound is 1), just increasing bound by the same range size is wanted (so adding MaxInputAmount achieves this) * Add logic to extend search bounds when finding optimal amount in - Increases max iterations to 17 to allow for situation when we need to increase the upper bound - Add new ExtendedMaxInputAmount variable for us to use as this new max bound range - Replace bound changing logic from iteratively changing the range to immediately giving our max range * Move range extension into it's own helper function * basic benchmark testing for posthandler and epoch hook * Add SwapAmountOut Test * Add extended range test - Tests binary search range extension logic works properly * Add panic catching test - This currently fails, this is on purpose so we don't merge the PR and think we are good until this passes and the panics are handled properly * dynamic step size * pool points only incremented if profitable * adding sanity checks for pool point calcs, nits * Update doomsday testing accounting for refund system - Makes pool 41 reserves to have an arb opportunity in the doomsday routes - Changes tx limit and block limit specifically for doomsday testing * Return nil for no profit opportunity * adding E2E, removing atom as a base denom, more testing for post handler * find max profit test fix * nit * backporting version tag to 14.x * comment update for protorev admin account --------- Co-authored-by: David Terpay <[email protected]> (cherry picked from commit 23b13a1) # Conflicts: # tests/e2e/configurer/chain/queries.go

* Protorev smarter logic and more testing (#4181) * Add generalized tests Added: - 4 pool mainnet route test - 2 pool mainnet route test - Non-osmo/atom denom test * Add stableswap doomsday test - Currently panics due to solveCFMMBinarySearchMulti in stableswap calculation * Add tests for pool point limits - Tests that the per tx limit works properly - Tests that the per block limit works properly * Add pre-check before binary search - Adds a pre-check before the binary search that tells us if we need to run the binary search at all - Reduces overall computation/time by avoiding binary searching profit amounts over routes without any profit opportunity * Add denom to make test pass - Added "test/3" denom to test non-osmo/atom denoms in previous PR - didn't notice it broke one test that expected us to only have 2 denoms, so adding "test/3" here so everything passes * Implement minimum change to have smarter binary search bounds - This implements a way for us to have a binary search bound that can increase in size to account for large trades above the default range - I believe this is the minimum-amount-of-new-code approach, but may not be the optimal approach (what this method avoids is having to save the amount in or amount out of the original swap, and then backtracking/converting that amount to the base denom for an arb, and creating bounds off of that) * Switch range increasing logic - Doubling isn't the wanted behavior (only wanted if lower bound is 1), just increasing bound by the same range size is wanted (so adding MaxInputAmount achieves this) * Add logic to extend search bounds when finding optimal amount in - Increases max iterations to 17 to allow for situation when we need to increase the upper bound - Add new ExtendedMaxInputAmount variable for us to use as this new max bound range - Replace bound changing logic from iteratively changing the range to immediately giving our max range * Move range extension into it's own helper function * basic benchmark testing for posthandler and epoch hook * Add SwapAmountOut Test * Add extended range test - Tests binary search range extension logic works properly * Add panic catching test - This currently fails, this is on purpose so we don't merge the PR and think we are good until this passes and the panics are handled properly * dynamic step size * pool points only incremented if profitable * adding sanity checks for pool point calcs, nits * Update doomsday testing accounting for refund system - Makes pool 41 reserves to have an arb opportunity in the doomsday routes - Changes tx limit and block limit specifically for doomsday testing * Return nil for no profit opportunity * adding E2E, removing atom as a base denom, more testing for post handler * find max profit test fix * nit * backporting version tag to 14.x * comment update for protorev admin account --------- Co-authored-by: David Terpay <[email protected]> (cherry picked from commit 23b13a1) # Conflicts: # tests/e2e/configurer/chain/queries.go * Fix merge conflict --------- Co-authored-by: Jeremy Liu <[email protected]> Co-authored-by: Dev Ojha <[email protected]>

* Add generalized tests Added: - 4 pool mainnet route test - 2 pool mainnet route test - Non-osmo/atom denom test * Add stableswap doomsday test - Currently panics due to solveCFMMBinarySearchMulti in stableswap calculation * Add tests for pool point limits - Tests that the per tx limit works properly - Tests that the per block limit works properly * Add pre-check before binary search - Adds a pre-check before the binary search that tells us if we need to run the binary search at all - Reduces overall computation/time by avoiding binary searching profit amounts over routes without any profit opportunity * Add denom to make test pass - Added "test/3" denom to test non-osmo/atom denoms in previous PR - didn't notice it broke one test that expected us to only have 2 denoms, so adding "test/3" here so everything passes * Implement minimum change to have smarter binary search bounds - This implements a way for us to have a binary search bound that can increase in size to account for large trades above the default range - I believe this is the minimum-amount-of-new-code approach, but may not be the optimal approach (what this method avoids is having to save the amount in or amount out of the original swap, and then backtracking/converting that amount to the base denom for an arb, and creating bounds off of that) * Switch range increasing logic - Doubling isn't the wanted behavior (only wanted if lower bound is 1), just increasing bound by the same range size is wanted (so adding MaxInputAmount achieves this) * Add logic to extend search bounds when finding optimal amount in - Increases max iterations to 17 to allow for situation when we need to increase the upper bound - Add new ExtendedMaxInputAmount variable for us to use as this new max bound range - Replace bound changing logic from iteratively changing the range to immediately giving our max range * Move range extension into it's own helper function * basic benchmark testing for posthandler and epoch hook * Add SwapAmountOut Test * Add extended range test - Tests binary search range extension logic works properly * Add panic catching test - This currently fails, this is on purpose so we don't merge the PR and think we are good until this passes and the panics are handled properly * dynamic step size * pool points only incremented if profitable * adding sanity checks for pool point calcs, nits * Update doomsday testing accounting for refund system - Makes pool 41 reserves to have an arb opportunity in the doomsday routes - Changes tx limit and block limit specifically for doomsday testing * Return nil for no profit opportunity * adding E2E, removing atom as a base denom, more testing for post handler * find max profit test fix * nit * backporting version tag to 14.x * comment update for protorev admin account --------- Co-authored-by: David Terpay <[email protected]>

NotJeremyLiu requested a review from davidterpay January 31, 2023 15:35

davidterpay added the V:state/breaking State machine breaking PR label Jan 31, 2023

NotJeremyLiu requested a review from bpiv400 February 1, 2023 14:27

davidterpay mentioned this pull request Feb 3, 2023

x/protorev PR Breakdown #3594

Closed

8 tasks

davidterpay force-pushed the protorev-smarter-logic-and-more-testing branch from 5ff4ed4 to 8a27d62 Compare February 13, 2023 17:14

NotJeremyLiu marked this pull request as ready for review February 14, 2023 15:02

NotJeremyLiu commented Feb 14, 2023

View reviewed changes

davidterpay reviewed Feb 15, 2023

View reviewed changes

NotJeremyLiu requested review from a team and ValarDragon February 15, 2023 23:17

p0mvn reviewed Feb 16, 2023

View reviewed changes

p0mvn mentioned this pull request Feb 16, 2023

chore(e2e): v14.x protorev genesis #4343

Merged

ValarDragon reviewed Feb 16, 2023

View reviewed changes

NotJeremyLiu and others added 12 commits February 16, 2023 11:19

Add generalized tests

0a194cc

Added: - 4 pool mainnet route test - 2 pool mainnet route test - Non-osmo/atom denom test

Add stableswap doomsday test

ded054f

- Currently panics due to solveCFMMBinarySearchMulti in stableswap calculation

Add tests for pool point limits

7b5d338

- Tests that the per tx limit works properly - Tests that the per block limit works properly

Add pre-check before binary search

736b1b6

- Adds a pre-check before the binary search that tells us if we need to run the binary search at all - Reduces overall computation/time by avoiding binary searching profit amounts over routes without any profit opportunity

Add denom to make test pass

9c579ed

- Added "test/3" denom to test non-osmo/atom denoms in previous PR - didn't notice it broke one test that expected us to only have 2 denoms, so adding "test/3" here so everything passes

Switch range increasing logic

1dcac42

- Doubling isn't the wanted behavior (only wanted if lower bound is 1), just increasing bound by the same range size is wanted (so adding MaxInputAmount achieves this)

Move range extension into it's own helper function

d9a0469

basic benchmark testing for posthandler and epoch hook

34e5320

Add SwapAmountOut Test

d9ab499

Add extended range test

fd4d371

- Tests binary search range extension logic works properly

NotJeremyLiu and others added 6 commits February 16, 2023 11:19

Add panic catching test

a49733e

- This currently fails, this is on purpose so we don't merge the PR and think we are good until this passes and the panics are handled properly

dynamic step size

29ef00f

pool points only incremented if profitable

03f863b

adding sanity checks for pool point calcs, nits

fe23f2f

Update doomsday testing accounting for refund system

6229875

- Makes pool 41 reserves to have an arb opportunity in the doomsday routes - Changes tx limit and block limit specifically for doomsday testing

Return nil for no profit opportunity

4620f66

ValarDragon reviewed Feb 16, 2023

View reviewed changes

davidterpay added 4 commits February 16, 2023 11:54

adding E2E, removing atom as a base denom, more testing for post handler

1007e6b

find max profit test fix

51c6249

nit

c9f6938

backporting version tag to 14.x

450fd52

davidterpay force-pushed the protorev-smarter-logic-and-more-testing branch from eb19820 to 450fd52 Compare February 16, 2023 16:55

ValarDragon reviewed Feb 16, 2023

View reviewed changes

tests/e2e/configurer/chain/queries.go Show resolved Hide resolved

davidterpay mentioned this pull request Feb 16, 2023

ProtoRev: Settings manager (Admin Account) configured in module genesis #4349

Closed

ValarDragon reviewed Feb 16, 2023

View reviewed changes

ValarDragon approved these changes Feb 16, 2023

View reviewed changes

ValarDragon added the A:backport/v15.x backport patches to v15.x branch label Feb 16, 2023

comment update for protorev admin account

5b7b595

ValarDragon merged commit 23b13a1 into main Feb 16, 2023

ValarDragon deleted the protorev-smarter-logic-and-more-testing branch February 16, 2023 18:21

mergify bot mentioned this pull request Feb 16, 2023

Protorev smarter logic and more testing (backport #4181) #4352

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Protorev smarter logic and more testing #4181

Protorev smarter logic and more testing #4181

NotJeremyLiu commented Jan 31, 2023 •

edited by davidterpay

Loading

NotJeremyLiu left a comment

NotJeremyLiu Feb 14, 2023

NotJeremyLiu Feb 15, 2023

NotJeremyLiu commented Feb 14, 2023 •

edited

Loading

davidterpay commented Feb 15, 2023

davidterpay Feb 15, 2023 •

edited

Loading

ValarDragon Feb 16, 2023

davidterpay Feb 15, 2023 •

edited

Loading

davidterpay Feb 15, 2023 •

edited

Loading

p0mvn Feb 16, 2023

ValarDragon Feb 16, 2023

ValarDragon Feb 16, 2023

davidterpay Feb 16, 2023

ValarDragon Feb 16, 2023

ValarDragon Feb 16, 2023 •

edited

Loading

davidterpay Feb 16, 2023

ValarDragon Feb 16, 2023

ValarDragon Feb 16, 2023

ValarDragon Feb 16, 2023

ValarDragon left a comment

		// Hot routes that are configured on genesis
		repeated TokenPairArbRoutes token_pairs = 2 [ (gogoproto.nullable) = false ];

	// Max iterations for binary search (log2(16_384) = 14)
	// Max iterations for binary search (log2(131_072) = 17)

Protorev smarter logic and more testing #4181

Protorev smarter logic and more testing #4181

Conversation

NotJeremyLiu commented Jan 31, 2023 • edited by davidterpay Loading

What is the purpose of the change

Brief Changelog

Testing and Verifying

NotJeremyLiu left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

NotJeremyLiu commented Feb 14, 2023 • edited Loading

davidterpay commented Feb 15, 2023

davidterpay Feb 15, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

davidterpay Feb 15, 2023 • edited Loading

Choose a reason for hiding this comment

davidterpay Feb 15, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ValarDragon Feb 16, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ValarDragon left a comment

Choose a reason for hiding this comment

NotJeremyLiu commented Jan 31, 2023 •

edited by davidterpay

Loading

NotJeremyLiu commented Feb 14, 2023 •

edited

Loading

davidterpay Feb 15, 2023 •

edited

Loading

davidterpay Feb 15, 2023 •

edited

Loading

davidterpay Feb 15, 2023 •

edited

Loading

ValarDragon Feb 16, 2023 •

edited

Loading