influxdb/sqlite/migrations/0008_migrate_remotes_null_r...

85 lines
2.7 KiB
SQL

-- Adds the "NOT NULL" to remote_org_id
ALTER TABLE remotes RENAME TO _remotes_old;
CREATE TABLE remotes (
id VARCHAR(16) NOT NULL PRIMARY KEY,
org_id VARCHAR(16) NOT NULL,
name TEXT NOT NULL,
description TEXT,
remote_url TEXT NOT NULL,
remote_api_token TEXT NOT NULL,
remote_org_id VARCHAR(16) NOT NULL,
allow_insecure_tls BOOLEAN NOT NULL,
created_at TIMESTAMP NOT NULL,
updated_at TIMESTAMP NOT NULL,
CONSTRAINT remotes_uniq_orgid_name UNIQUE (org_id, name)
);
INSERT INTO remotes (
id,
org_id,
name,
description,
remote_url,
remote_api_token,
remote_org_id,
allow_insecure_tls,
created_at,
updated_at
) SELECT * FROM _remotes_old WHERE remote_org_id IS NOT NULL;
-- Edit the replications table as the remotes table key has changed
ALTER TABLE replications RENAME TO _replications_old;
CREATE TABLE replications
(
id VARCHAR(16) NOT NULL PRIMARY KEY,
org_id VARCHAR(16) NOT NULL,
name TEXT NOT NULL,
description TEXT,
remote_id VARCHAR(16) NOT NULL,
local_bucket_id VARCHAR(16) NOT NULL,
remote_bucket_id VARCHAR(16),
remote_bucket_name TEXT DEFAULT '',
max_queue_size_bytes INTEGER NOT NULL,
max_age_seconds INTEGER NOT NULL,
latest_response_code INTEGER,
latest_error_message TEXT,
drop_non_retryable_data BOOLEAN NOT NULL,
created_at TIMESTAMP NOT NULL,
updated_at TIMESTAMP NOT NULL,
CONSTRAINT replications_uniq_orgid_name UNIQUE (org_id, name),
CONSTRAINT replications_one_of_id_name CHECK (remote_bucket_id IS NOT NULL OR remote_bucket_name != '')
);
INSERT INTO replications (
id,
org_id,
name,
description,
remote_id,
local_bucket_id,
remote_bucket_id,
remote_bucket_name,
max_queue_size_bytes,
max_age_seconds,
latest_response_code,
latest_error_message,
drop_non_retryable_data,
created_at,
updated_at
) SELECT * FROM _replications_old;
DROP TABLE _replications_old;
DROP TABLE _remotes_old;
-- The DROP _remotes has to be at the end due to the FK from replications remote_id to remotes id.
-- The replications table will follow the ALTER TABLE and FK to _remotes until we
-- reinsert. By putting the DROP after all the data is re-entered, it will stay consistent throughout the process.
-- Create indexes on lookup patterns we expect to be common
CREATE INDEX idx_local_bucket_id_per_org ON replications (org_id, local_bucket_id);
-- Create indexes on lookup patterns we expect to be common
CREATE INDEX idx_remote_url_per_org ON remotes (org_id, remote_url);