Merge pull request #38 from vulcanize/migrations

consolidate migrations
This commit is contained in:
Ian Norden 2021-03-17 16:47:32 -05:00 committed by GitHub
commit df454c414e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
16 changed files with 131 additions and 296 deletions

View File

@ -5,7 +5,8 @@ CREATE TABLE nodes (
genesis_block VARCHAR(66), genesis_block VARCHAR(66),
network_id VARCHAR, network_id VARCHAR,
node_id VARCHAR(128), node_id VARCHAR(128),
CONSTRAINT node_uc UNIQUE (genesis_block, network_id, node_id) chain_id INTEGER DEFAULT 1,
CONSTRAINT node_uc UNIQUE (genesis_block, network_id, node_id, chain_id)
); );
-- +goose Down -- +goose Down

View File

@ -8,7 +8,6 @@ CREATE TABLE eth.transaction_cids (
mh_key TEXT NOT NULL REFERENCES public.blocks (key) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED, mh_key TEXT NOT NULL REFERENCES public.blocks (key) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED,
dst VARCHAR(66) NOT NULL, dst VARCHAR(66) NOT NULL,
src VARCHAR(66) NOT NULL, src VARCHAR(66) NOT NULL,
deployment BOOL NOT NULL,
tx_data BYTEA, tx_data BYTEA,
UNIQUE (header_id, tx_hash) UNIQUE (header_id, tx_hash)
); );

View File

@ -11,6 +11,8 @@ CREATE TABLE eth.receipt_cids (
topic2s VARCHAR(66)[], topic2s VARCHAR(66)[],
topic3s VARCHAR(66)[], topic3s VARCHAR(66)[],
log_contracts VARCHAR(66)[], log_contracts VARCHAR(66)[],
post_state VARCHAR(66),
post_status INTEGER,
UNIQUE (tx_id) UNIQUE (tx_id)
); );

View File

@ -1,6 +1,6 @@
-- +goose Up -- +goose Up
CREATE TABLE eth.state_cids ( CREATE TABLE eth.state_cids (
id SERIAL PRIMARY KEY, id BIGSERIAL PRIMARY KEY,
header_id INTEGER NOT NULL REFERENCES eth.header_cids (id) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED, header_id INTEGER NOT NULL REFERENCES eth.header_cids (id) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED,
state_leaf_key VARCHAR(66), state_leaf_key VARCHAR(66),
cid TEXT NOT NULL, cid TEXT NOT NULL,

View File

@ -1,7 +1,7 @@
-- +goose Up -- +goose Up
CREATE TABLE eth.storage_cids ( CREATE TABLE eth.storage_cids (
id SERIAL PRIMARY KEY, id BIGSERIAL PRIMARY KEY,
state_id INTEGER NOT NULL REFERENCES eth.state_cids (id) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED, state_id BIGINT NOT NULL REFERENCES eth.state_cids (id) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED,
storage_leaf_key VARCHAR(66), storage_leaf_key VARCHAR(66),
cid TEXT NOT NULL, cid TEXT NOT NULL,
mh_key TEXT NOT NULL REFERENCES public.blocks (key) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED, mh_key TEXT NOT NULL REFERENCES public.blocks (key) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED,

View File

@ -1,7 +1,7 @@
-- +goose Up -- +goose Up
CREATE TABLE eth.state_accounts ( CREATE TABLE eth.state_accounts (
id SERIAL PRIMARY KEY, id SERIAL PRIMARY KEY,
state_id INTEGER NOT NULL REFERENCES eth.state_cids (id) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED, state_id BIGINT NOT NULL REFERENCES eth.state_cids (id) ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED,
balance NUMERIC NOT NULL, balance NUMERIC NOT NULL,
nonce INTEGER NOT NULL, nonce INTEGER NOT NULL,
code_hash BYTEA NOT NULL, code_hash BYTEA NOT NULL,

View File

@ -1,21 +0,0 @@
-- +goose Up
ALTER TABLE public.nodes
ADD COLUMN chain_id INTEGER DEFAULT 1;
ALTER TABLE public.nodes
DROP CONSTRAINT node_uc;
ALTER TABLE public.nodes
ADD CONSTRAINT node_uc
UNIQUE (genesis_block, network_id, node_id, chain_id);
-- +goose Down
ALTER TABLE public.nodes
DROP CONSTRAINT node_uc;
ALTER TABLE public.nodes
ADD CONSTRAINT node_uc
UNIQUE (genesis_block, network_id, node_id);
ALTER TABLE public.nodes
DROP COLUMN chain_id;

View File

@ -1,4 +1,39 @@
-- +goose Up -- +goose Up
-- +goose StatementBegin
-- returns if a storage node at the provided path was removed in the range > the provided height and <= the provided block hash
CREATE OR REPLACE FUNCTION was_storage_removed(path BYTEA, height BIGINT, hash VARCHAR(66)) RETURNS BOOLEAN
AS $$
SELECT exists(SELECT 1
FROM eth.storage_cids
INNER JOIN eth.state_cids ON (storage_cids.state_id = state_cids.id)
INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id)
WHERE storage_path = path
AND block_number > height
AND block_number <= (SELECT block_number
FROM eth.header_cids
WHERE block_hash = hash)
AND storage_cids.node_type = 3
LIMIT 1);
$$ LANGUAGE SQL;
-- +goose StatementEnd
-- +goose StatementBegin
-- returns if a state node at the provided path was removed in the range > the provided height and <= the provided block hash
CREATE OR REPLACE FUNCTION was_state_removed(path BYTEA, height BIGINT, hash VARCHAR(66)) RETURNS BOOLEAN
AS $$
SELECT exists(SELECT 1
FROM eth.state_cids
INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id)
WHERE state_path = path
AND block_number > height
AND block_number <= (SELECT block_number
FROM eth.header_cids
WHERE block_hash = hash)
AND state_cids.node_type = 3
LIMIT 1);
$$ LANGUAGE SQL;
-- +goose StatementEnd
-- +goose StatementBegin -- +goose StatementBegin
CREATE TYPE child_result AS ( CREATE TYPE child_result AS (
has_child BOOLEAN, has_child BOOLEAN,
@ -8,27 +43,27 @@ CREATE TYPE child_result AS (
CREATE OR REPLACE FUNCTION has_child(hash VARCHAR(66), height BIGINT) RETURNS child_result AS CREATE OR REPLACE FUNCTION has_child(hash VARCHAR(66), height BIGINT) RETURNS child_result AS
$BODY$ $BODY$
DECLARE DECLARE
child_height INT; child_height INT;
temp_child eth.header_cids; temp_child eth.header_cids;
new_child_result child_result; new_child_result child_result;
BEGIN BEGIN
child_height = height + 1; child_height = height + 1;
-- short circuit if there are no children -- short circuit if there are no children
SELECT exists(SELECT 1 SELECT exists(SELECT 1
FROM eth.header_cids FROM eth.header_cids
WHERE parent_hash = hash WHERE parent_hash = hash
AND block_number = child_height AND block_number = child_height
LIMIT 1) LIMIT 1)
INTO new_child_result.has_child; INTO new_child_result.has_child;
-- collect all the children for this header -- collect all the children for this header
IF new_child_result.has_child THEN IF new_child_result.has_child THEN
FOR temp_child IN FOR temp_child IN
SELECT * FROM eth.header_cids WHERE parent_hash = hash AND block_number = child_height SELECT * FROM eth.header_cids WHERE parent_hash = hash AND block_number = child_height
LOOP LOOP
new_child_result.children = array_append(new_child_result.children, temp_child); new_child_result.children = array_append(new_child_result.children, temp_child);
END LOOP; END LOOP;
END IF; END IF;
RETURN new_child_result; RETURN new_child_result;
END END
$BODY$ $BODY$
LANGUAGE 'plpgsql'; LANGUAGE 'plpgsql';
@ -38,7 +73,7 @@ LANGUAGE 'plpgsql';
CREATE OR REPLACE FUNCTION canonical_header_from_array(headers eth.header_cids[]) RETURNS eth.header_cids AS CREATE OR REPLACE FUNCTION canonical_header_from_array(headers eth.header_cids[]) RETURNS eth.header_cids AS
$BODY$ $BODY$
DECLARE DECLARE
canonical_header eth.header_cids; canonical_header eth.header_cids;
canonical_child eth.header_cids; canonical_child eth.header_cids;
header eth.header_cids; header eth.header_cids;
current_child_result child_result; current_child_result child_result;
@ -57,25 +92,25 @@ BEGIN
current_header_with_child = header; current_header_with_child = header;
-- and add the children to the growing set of child headers -- and add the children to the growing set of child headers
child_headers = array_cat(child_headers, current_child_result.children); child_headers = array_cat(child_headers, current_child_result.children);
END IF; END IF;
END LOOP; END LOOP;
-- if none of the headers had children, none is more canonical than the other -- if none of the headers had children, none is more canonical than the other
IF has_children_count = 0 THEN IF has_children_count = 0 THEN
-- return the first one selected -- return the first one selected
SELECT * INTO canonical_header FROM unnest(headers) LIMIT 1; SELECT * INTO canonical_header FROM unnest(headers) LIMIT 1;
-- if only one header had children, it can be considered the heaviest/canonical header of the set -- if only one header had children, it can be considered the heaviest/canonical header of the set
ELSIF has_children_count = 1 THEN ELSIF has_children_count = 1 THEN
-- return the only header with a child -- return the only header with a child
canonical_header = current_header_with_child; canonical_header = current_header_with_child;
-- if there are multiple headers with children -- if there are multiple headers with children
ELSE ELSE
-- find the canonical header from the child set -- find the canonical header from the child set
canonical_child = canonical_header_from_array(child_headers); canonical_child = canonical_header_from_array(child_headers);
-- the header that is parent to this header, is the canonical header at this level -- the header that is parent to this header, is the canonical header at this level
SELECT * INTO canonical_header FROM unnest(headers) SELECT * INTO canonical_header FROM unnest(headers)
WHERE block_hash = canonical_child.parent_hash; WHERE block_hash = canonical_child.parent_hash;
END IF; END IF;
RETURN canonical_header; RETURN canonical_header;
END END
$BODY$ $BODY$
LANGUAGE 'plpgsql'; LANGUAGE 'plpgsql';
@ -85,17 +120,17 @@ LANGUAGE 'plpgsql';
CREATE OR REPLACE FUNCTION canonical_header_id(height BIGINT) RETURNS INTEGER AS CREATE OR REPLACE FUNCTION canonical_header_id(height BIGINT) RETURNS INTEGER AS
$BODY$ $BODY$
DECLARE DECLARE
canonical_header eth.header_cids; canonical_header eth.header_cids;
headers eth.header_cids[]; headers eth.header_cids[];
header_count INT; header_count INT;
temp_header eth.header_cids; temp_header eth.header_cids;
BEGIN BEGIN
-- collect all headers at this height -- collect all headers at this height
FOR temp_header IN FOR temp_header IN
SELECT * FROM eth.header_cids WHERE block_number = height SELECT * FROM eth.header_cids WHERE block_number = height
LOOP LOOP
headers = array_append(headers, temp_header); headers = array_append(headers, temp_header);
END LOOP; END LOOP;
-- count the number of headers collected -- count the number of headers collected
header_count = array_length(headers, 1); header_count = array_length(headers, 1);
-- if we have less than 1 header, return NULL -- if we have less than 1 header, return NULL
@ -105,16 +140,18 @@ BEGIN
ELSIF header_count = 1 THEN ELSIF header_count = 1 THEN
RETURN headers[1].id; RETURN headers[1].id;
-- if we have multiple headers we need to determine which one is canonical -- if we have multiple headers we need to determine which one is canonical
ELSE ELSE
canonical_header = canonical_header_from_array(headers); canonical_header = canonical_header_from_array(headers);
RETURN canonical_header.id; RETURN canonical_header.id;
END IF; END IF;
END; END;
$BODY$ $BODY$
LANGUAGE 'plpgsql'; LANGUAGE 'plpgsql';
-- +goose StatementEnd -- +goose StatementEnd
-- +goose Down -- +goose Down
DROP FUNCTION was_storage_removed;
DROP FUNCTION was_state_removed;
DROP FUNCTION canonical_header_id; DROP FUNCTION canonical_header_id;
DROP FUNCTION canonical_header_from_array; DROP FUNCTION canonical_header_from_array;
DROP FUNCTION has_child; DROP FUNCTION has_child;

View File

@ -1,48 +0,0 @@
-- +goose Up
-- +goose StatementBegin
-- returns the number of child headers that reference backwards to the header with the provided hash
CREATE OR REPLACE FUNCTION header_weight(hash VARCHAR(66)) RETURNS BIGINT
AS $$
WITH RECURSIVE validator AS (
SELECT block_hash, parent_hash, block_number
FROM eth.header_cids
WHERE block_hash = hash
UNION
SELECT eth.header_cids.block_hash, eth.header_cids.parent_hash, eth.header_cids.block_number
FROM eth.header_cids
INNER JOIN validator
ON eth.header_cids.parent_hash = validator.block_hash
AND eth.header_cids.block_number = validator.block_number + 1
)
SELECT COUNT(*) FROM validator;
$$ LANGUAGE SQL;
-- +goose StatementEnd
-- +goose StatementBegin
-- returns the id for the header at the provided height which is heaviest
CREATE OR REPLACE FUNCTION canonical_header(height BIGINT) RETURNS INT AS
$BODY$
DECLARE
current_weight INT;
heaviest_weight INT DEFAULT 0;
heaviest_id INT;
r eth.header_cids%ROWTYPE;
BEGIN
FOR r IN SELECT * FROM eth.header_cids
WHERE block_number = height
LOOP
SELECT INTO current_weight * FROM header_weight(r.block_hash);
IF current_weight > heaviest_weight THEN
heaviest_weight := current_weight;
heaviest_id := r.id;
END IF;
END LOOP;
RETURN heaviest_id;
END
$BODY$
LANGUAGE 'plpgsql';
-- +goose StatementEnd
-- +goose Down
DROP FUNCTION header_weight;
DROP FUNCTION canonical_header;

View File

@ -1,7 +0,0 @@
-- +goose Up
ALTER TABLE eth.transaction_cids
DROP COLUMN deployment;
-- +goose Down
ALTER TABLE eth.transaction_cids
ADD COLUMN deployment BOOL NOT NULL DEFAULT FALSE;

View File

@ -1,25 +0,0 @@
-- +goose Up
ALTER TABLE eth.storage_cids
ALTER COLUMN state_id TYPE BIGINT;
ALTER TABLE eth.state_accounts
ALTER COLUMN state_id TYPE BIGINT;
ALTER TABLE eth.state_cids
ALTER COLUMN id TYPE BIGINT;
ALTER TABLE eth.storage_cids
ALTER COLUMN id TYPE BIGINT;
-- +goose Down
ALTER TABLE eth.storage_cids
ALTER COLUMN id TYPE INTEGER;
ALTER TABLE eth.state_cids
ALTER COLUMN id TYPE INTEGER;
ALTER TABLE eth.state_accounts
ALTER COLUMN state_id TYPE INTEGER;
ALTER TABLE eth.storage_cids
ALTER COLUMN state_id TYPE INTEGER;

View File

@ -1,39 +0,0 @@
-- +goose Up
-- +goose StatementBegin
-- returns if a storage node at the provided path was removed in the range > the provided height and <= the provided block hash
CREATE OR REPLACE FUNCTION was_storage_removed(path BYTEA, height BIGINT, hash VARCHAR(66)) RETURNS BOOLEAN
AS $$
SELECT exists(SELECT 1
FROM eth.storage_cids
INNER JOIN eth.state_cids ON (storage_cids.state_id = state_cids.id)
INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id)
WHERE storage_path = path
AND block_number > height
AND block_number <= (SELECT block_number
FROM eth.header_cids
WHERE block_hash = hash)
AND storage_cids.node_type = 3
LIMIT 1);
$$ LANGUAGE SQL;
-- +goose StatementEnd
-- +goose StatementBegin
-- returns if a state node at the provided path was removed in the range > the provided height and <= the provided block hash
CREATE OR REPLACE FUNCTION was_state_removed(path BYTEA, height BIGINT, hash VARCHAR(66)) RETURNS BOOLEAN
AS $$
SELECT exists(SELECT 1
FROM eth.state_cids
INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id)
WHERE state_path = path
AND block_number > height
AND block_number <= (SELECT block_number
FROM eth.header_cids
WHERE block_hash = hash)
AND state_cids.node_type = 3
LIMIT 1);
$$ LANGUAGE SQL;
-- +goose StatementEnd
-- +goose Down
DROP FUNCTION was_storage_removed;
DROP FUNCTION was_state_removed;

View File

@ -1,13 +0,0 @@
-- +goose Up
ALTER TABLE eth.receipt_cids
ADD COLUMN post_state VARCHAR(66);
ALTER TABLE eth.receipt_cids
ADD COLUMN post_status INTEGER;
-- +goose Down
ALTER TABLE eth.receipt_cids
DROP COLUMN post_status;
ALTER TABLE eth.receipt_cids
DROP COLUMN post_state;

View File

@ -103,33 +103,6 @@ end;
$_$; $_$;
--
-- Name: canonical_header(bigint); Type: FUNCTION; Schema: public; Owner: -
--
CREATE FUNCTION public.canonical_header(height bigint) RETURNS integer
LANGUAGE plpgsql
AS $$
DECLARE
current_weight INT;
heaviest_weight INT DEFAULT 0;
heaviest_id INT;
r eth.header_cids%ROWTYPE;
BEGIN
FOR r IN SELECT * FROM eth.header_cids
WHERE block_number = height
LOOP
SELECT INTO current_weight * FROM header_weight(r.block_hash);
IF current_weight > heaviest_weight THEN
heaviest_weight := current_weight;
heaviest_id := r.id;
END IF;
END LOOP;
RETURN heaviest_id;
END
$$;
-- --
-- Name: canonical_header_from_array(eth.header_cids[]); Type: FUNCTION; Schema: public; Owner: - -- Name: canonical_header_from_array(eth.header_cids[]); Type: FUNCTION; Schema: public; Owner: -
-- --
@ -138,7 +111,7 @@ CREATE FUNCTION public.canonical_header_from_array(headers eth.header_cids[]) RE
LANGUAGE plpgsql LANGUAGE plpgsql
AS $$ AS $$
DECLARE DECLARE
canonical_header eth.header_cids; canonical_header eth.header_cids;
canonical_child eth.header_cids; canonical_child eth.header_cids;
header eth.header_cids; header eth.header_cids;
current_child_result child_result; current_child_result child_result;
@ -157,25 +130,25 @@ BEGIN
current_header_with_child = header; current_header_with_child = header;
-- and add the children to the growing set of child headers -- and add the children to the growing set of child headers
child_headers = array_cat(child_headers, current_child_result.children); child_headers = array_cat(child_headers, current_child_result.children);
END IF; END IF;
END LOOP; END LOOP;
-- if none of the headers had children, none is more canonical than the other -- if none of the headers had children, none is more canonical than the other
IF has_children_count = 0 THEN IF has_children_count = 0 THEN
-- return the first one selected -- return the first one selected
SELECT * INTO canonical_header FROM unnest(headers) LIMIT 1; SELECT * INTO canonical_header FROM unnest(headers) LIMIT 1;
-- if only one header had children, it can be considered the heaviest/canonical header of the set -- if only one header had children, it can be considered the heaviest/canonical header of the set
ELSIF has_children_count = 1 THEN ELSIF has_children_count = 1 THEN
-- return the only header with a child -- return the only header with a child
canonical_header = current_header_with_child; canonical_header = current_header_with_child;
-- if there are multiple headers with children -- if there are multiple headers with children
ELSE ELSE
-- find the canonical header from the child set -- find the canonical header from the child set
canonical_child = canonical_header_from_array(child_headers); canonical_child = canonical_header_from_array(child_headers);
-- the header that is parent to this header, is the canonical header at this level -- the header that is parent to this header, is the canonical header at this level
SELECT * INTO canonical_header FROM unnest(headers) SELECT * INTO canonical_header FROM unnest(headers)
WHERE block_hash = canonical_child.parent_hash; WHERE block_hash = canonical_child.parent_hash;
END IF; END IF;
RETURN canonical_header; RETURN canonical_header;
END END
$$; $$;
@ -188,17 +161,17 @@ CREATE FUNCTION public.canonical_header_id(height bigint) RETURNS integer
LANGUAGE plpgsql LANGUAGE plpgsql
AS $$ AS $$
DECLARE DECLARE
canonical_header eth.header_cids; canonical_header eth.header_cids;
headers eth.header_cids[]; headers eth.header_cids[];
header_count INT; header_count INT;
temp_header eth.header_cids; temp_header eth.header_cids;
BEGIN BEGIN
-- collect all headers at this height -- collect all headers at this height
FOR temp_header IN FOR temp_header IN
SELECT * FROM eth.header_cids WHERE block_number = height SELECT * FROM eth.header_cids WHERE block_number = height
LOOP LOOP
headers = array_append(headers, temp_header); headers = array_append(headers, temp_header);
END LOOP; END LOOP;
-- count the number of headers collected -- count the number of headers collected
header_count = array_length(headers, 1); header_count = array_length(headers, 1);
-- if we have less than 1 header, return NULL -- if we have less than 1 header, return NULL
@ -208,10 +181,10 @@ BEGIN
ELSIF header_count = 1 THEN ELSIF header_count = 1 THEN
RETURN headers[1].id; RETURN headers[1].id;
-- if we have multiple headers we need to determine which one is canonical -- if we have multiple headers we need to determine which one is canonical
ELSE ELSE
canonical_header = canonical_header_from_array(headers); canonical_header = canonical_header_from_array(headers);
RETURN canonical_header.id; RETURN canonical_header.id;
END IF; END IF;
END; END;
$$; $$;
@ -224,53 +197,31 @@ CREATE FUNCTION public.has_child(hash character varying, height bigint) RETURNS
LANGUAGE plpgsql LANGUAGE plpgsql
AS $$ AS $$
DECLARE DECLARE
child_height INT; child_height INT;
temp_child eth.header_cids; temp_child eth.header_cids;
new_child_result child_result; new_child_result child_result;
BEGIN BEGIN
child_height = height + 1; child_height = height + 1;
-- short circuit if there are no children -- short circuit if there are no children
SELECT exists(SELECT 1 SELECT exists(SELECT 1
FROM eth.header_cids FROM eth.header_cids
WHERE parent_hash = hash WHERE parent_hash = hash
AND block_number = child_height AND block_number = child_height
LIMIT 1) LIMIT 1)
INTO new_child_result.has_child; INTO new_child_result.has_child;
-- collect all the children for this header -- collect all the children for this header
IF new_child_result.has_child THEN IF new_child_result.has_child THEN
FOR temp_child IN FOR temp_child IN
SELECT * FROM eth.header_cids WHERE parent_hash = hash AND block_number = child_height SELECT * FROM eth.header_cids WHERE parent_hash = hash AND block_number = child_height
LOOP LOOP
new_child_result.children = array_append(new_child_result.children, temp_child); new_child_result.children = array_append(new_child_result.children, temp_child);
END LOOP; END LOOP;
END IF; END IF;
RETURN new_child_result; RETURN new_child_result;
END END
$$; $$;
--
-- Name: header_weight(character varying); Type: FUNCTION; Schema: public; Owner: -
--
CREATE FUNCTION public.header_weight(hash character varying) RETURNS bigint
LANGUAGE sql
AS $$
WITH RECURSIVE validator AS (
SELECT block_hash, parent_hash, block_number
FROM eth.header_cids
WHERE block_hash = hash
UNION
SELECT eth.header_cids.block_hash, eth.header_cids.parent_hash, eth.header_cids.block_number
FROM eth.header_cids
INNER JOIN validator
ON eth.header_cids.parent_hash = validator.block_hash
AND eth.header_cids.block_number = validator.block_number + 1
)
SELECT COUNT(*) FROM validator;
$$;
-- --
-- Name: was_state_removed(bytea, bigint, character varying); Type: FUNCTION; Schema: public; Owner: - -- Name: was_state_removed(bytea, bigint, character varying); Type: FUNCTION; Schema: public; Owner: -
-- --
@ -278,7 +229,7 @@ $$;
CREATE FUNCTION public.was_state_removed(path bytea, height bigint, hash character varying) RETURNS boolean CREATE FUNCTION public.was_state_removed(path bytea, height bigint, hash character varying) RETURNS boolean
LANGUAGE sql LANGUAGE sql
AS $$ AS $$
SELECT exists(SELECT 1 SELECT exists(SELECT 1
FROM eth.state_cids FROM eth.state_cids
INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id) INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id)
WHERE state_path = path WHERE state_path = path
@ -298,7 +249,7 @@ $$;
CREATE FUNCTION public.was_storage_removed(path bytea, height bigint, hash character varying) RETURNS boolean CREATE FUNCTION public.was_storage_removed(path bytea, height bigint, hash character varying) RETURNS boolean
LANGUAGE sql LANGUAGE sql
AS $$ AS $$
SELECT exists(SELECT 1 SELECT exists(SELECT 1
FROM eth.storage_cids FROM eth.storage_cids
INNER JOIN eth.state_cids ON (storage_cids.state_id = state_cids.id) INNER JOIN eth.state_cids ON (storage_cids.state_id = state_cids.id)
INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id) INNER JOIN eth.header_cids ON (state_cids.header_id = header_cids.id)
@ -428,7 +379,6 @@ CREATE TABLE eth.state_cids (
-- --
CREATE SEQUENCE eth.state_cids_id_seq CREATE SEQUENCE eth.state_cids_id_seq
AS integer
START WITH 1 START WITH 1
INCREMENT BY 1 INCREMENT BY 1
NO MINVALUE NO MINVALUE
@ -464,7 +414,6 @@ CREATE TABLE eth.storage_cids (
-- --
CREATE SEQUENCE eth.storage_cids_id_seq CREATE SEQUENCE eth.storage_cids_id_seq
AS integer
START WITH 1 START WITH 1
INCREMENT BY 1 INCREMENT BY 1
NO MINVALUE NO MINVALUE