[experimenting] Add Balancer V2 indexer DB migrations

database/sql-balancer-indexer/README.md

@@ -0,0 +1,77 @@
+# Balancer-indexer migrations
+
+Flyway migrations for the Balancer V2 indexer's own per-network database (e.g.
+`mainnet_balancer_indexer`), kept out of the shared `../sql/` set so they don't
+run against the autopilot/orderbook main DBs. The local/e2e DB is
+`balancer_indexer`.
+
+The migration image ships every dir; init containers pick one via `-locations`:
+
+| DB                  | location                                              |
+|---------------------|-------------------------------------------------------|
+| autopilot/orderbook | `/flyway/sql` (default)                               |
+| pool-indexer        | `-locations=filesystem:/flyway/sql-pool-indexer`      |
+| balancer-indexer    | `-locations=filesystem:/flyway/sql-balancer-indexer`  |
+
+New Balancer-indexer migrations go here, never in `../sql/`. Unlike the
+pool-indexer's `V110` (duplicated from `../sql/` and cancelled there by
+`../sql/V111`), these tables are new to this directory, so there's no
+shared-set copy to cancel.
+
+## Schema
+
+The tables below live in the indexer's own per-network database, created by the
+migrations in this directory. The indexer stores *discovery* metadata only;
+dynamic state (balances, amplification, LBP weights, scaling factors, swap fee,
+paused) stays on-chain and is fetched by the driver at query time.
+
+### balancer\_v2\_checkpoints
+
+Highest finalized block processed per `factory_address` by the Balancer
+indexer. The indexer runs one process per network against its own DB, so
+there's no `chain_id` column.
+
+ Column            | Type   | Nullable | Details
+-------------------|--------|----------|--------
+ factory\_address  | bytea  | not null | Factory address (20 bytes)
+ block\_number     | bigint | not null |
+
+Indexes:
+- PRIMARY KEY: btree (`factory_address`)
+
+### balancer\_v2\_pools
+
+One row per registered pool, discovered from each factory's `PoolCreated`
+event. `pool_type` is derived from which factory created the pool (the
+factory→type map comes from config — no on-chain classification call). The
+weighted V0 vs V3-plus distinction is recovered driver-side from `factory`, so
+it isn't a separate `pool_type`.
+
+ Column          | Type   | Nullable | Details
+-----------------|--------|----------|--------
+ pool\_id        | bytea  | not null | 32-byte Balancer poolId
+ address         | bytea  | not null | Pool address (poolId's first 20 bytes)
+ factory         | bytea  | not null | Address of the factory that emitted `PoolCreated`
+ pool\_type      | text   | not null | `Weighted` \| `Stable` \| `ComposableStable` \| `LiquidityBootstrapping`. `CHECK`ed; stored as the string the API serves.
+ created\_block  | bigint | not null | Block in which the pool was created on-chain
+
+Indexes:
+- PRIMARY KEY: btree (`pool_id`)
+
+### balancer\_v2\_pool\_tokens
+
+Tokens per pool, in `Vault.getPoolTokens` order (`position`). `decimals` is
+nullable and filled in by the backfill task. `weight` is the Balancer Bfp
+(1e18 fixed-point) normalized weight, set only for weighted pools.
+
+ Column     | Type     | Nullable | Details
+------------|----------|----------|--------
+ pool\_id   | bytea    | not null | FK → `balancer_v2_pools(pool_id)`
+ position   | int      | not null | Token index within the pool (`getPoolTokens` order)
+ token      | bytea    | not null | Token address (20 bytes)
+ decimals   | smallint | nullable | `NULL` = not yet fetched. `-1` = sentinel for "fetched but call failed"
+ weight     | numeric  | nullable | Bfp (1e18) normalized weight; `NULL` for stable/composable-stable/LBP
+
+Indexes:
+- PRIMARY KEY: btree (`pool_id`, `position`)
+- Partial index on `(token)` with predicate `decimals IS NULL` to power the backfill scan.

database/sql-balancer-indexer/V001__balancer_v2.sql

@@ -0,0 +1,40 @@
+-- Tracks the highest finalized block fully processed per factory contract.
+-- A DB instance hosts a single network, so no `chain_id` column is needed.
+CREATE TABLE balancer_v2_checkpoints (
+    factory_address  BYTEA  NOT NULL,   -- factory address
+    block_number     BIGINT NOT NULL,
+    PRIMARY KEY (factory_address)
+);
+
+-- One row per registered pool, discovered from each factory's `PoolCreated`
+-- event. `pool_type` is derived from which factory created the pool (no
+-- on-chain classification); the weighted V0/V3-plus distinction is recovered
+-- from `factory`, so it isn't a separate type. `pool_type` is stored as the
+-- string the API serves so there's no int<->enum mapping at the boundary.
+CREATE TABLE balancer_v2_pools (
+    pool_id        BYTEA  NOT NULL,   -- 32-byte Balancer poolId
+    address        BYTEA  NOT NULL,   -- pool address (poolId's first 20 bytes)
+    factory        BYTEA  NOT NULL,
+    pool_type      TEXT   NOT NULL CHECK (pool_type IN ('Weighted', 'Stable', 'ComposableStable', 'LiquidityBootstrapping')),
+    created_block  BIGINT NOT NULL,
+    PRIMARY KEY (pool_id)
+);
+
+-- Tokens per pool, in `Vault.getPoolTokens` order (`position`). `decimals` is
+-- nullable and filled in by the backfill task. `weight` is the Balancer Bfp
+-- (1e18 fixed-point) normalized weight, set only for weighted pools; NULL for
+-- stable/composable-stable/LBP (their weights are absent or fetched on-chain).
+CREATE TABLE balancer_v2_pool_tokens (
+    pool_id   BYTEA    NOT NULL,
+    position  INT      NOT NULL,
+    token     BYTEA    NOT NULL,
+    decimals  SMALLINT,             -- NULL = not yet fetched; -1 = fetched but call failed
+    weight    NUMERIC,              -- Bfp (1e18); weighted pools only, else NULL
+    PRIMARY KEY (pool_id, position),
+    FOREIGN KEY (pool_id) REFERENCES balancer_v2_pools(pool_id)
+);
+
+-- Decimals backfill hot path. Partial on `IS NULL` so the index shrinks to
+-- near-empty once most rows are populated (real value or the `-1` "tried,
+-- failed" sentinel).
+CREATE INDEX ON balancer_v2_pool_tokens (token) WHERE decimals IS NULL;