Refactor inventory and module plan

Prepared before implementation. The frozen reference is authoritative.

Parameters and globals

OCF parameter	Default	Runtime global
`service_user`	`david`	`SERVICE_USER`
`service_name`	`postgres.service`	`SERVICE_NAME`
`container_name`	`postgres`	`CONTAINER_NAME`
`pg_os_user`	`postgres`	`PG_OS_USER`
`db_user`	`david`	`DB_USER`
`db_name`	`david`	`DB_NAME`
`rejoin_source_host`	``	`REJOIN_SOURCE_HOST`
`rejoin_source_port`	`5432`	`REJOIN_SOURCE_PORT`
`replication_user`	`replicator`	`REPLICATION_USER`
`replication_slot`	``	`REPLICATION_SLOT`
`application_name`	``	`APPLICATION_NAME`
`rejoin_allow_basebackup`	`false`	`REJOIN_ALLOW_BASEBACKUP`

application_name falls back to hostname -s. Each parameter has an OCF_RESKEY input and an OCF_RESKEY_*_default global. Other framework globals: OCF_ROOT, OCF_FUNCTIONS_DIR, OCF_RUNNING_PROMOTED (normalized from OCF_RUNNING_MASTER), OCF_SUCCESS, OCF_NOT_RUNNING, OCF_ERR_GENERIC, OCF_ERR_CONFIGURED, OCF_ERR_INSTALLED, OCF_ERR_UNIMPLEMENTED, OCF_RESOURCE_INSTANCE. The validate_all loop also assigns the existing nonlocal variable cmd.

Notification inputs: OCF_RESKEY_CRM_meta_notify_{type,operation,promote_uname,demote_uname,promoted_uname,unpromoted_uname}. State persists under /run/loxthelion; demote accepts matching pre-demote state aged 0–120 seconds.

Container inputs: PGDATA, POSTGRES_REPLICATION_PASSWORD, POSTGRES_REPLICATION_USER, POSTGRES_HA_SLOTS. Container command locals include BINDIR, blocks and the explicitly injected RA_PRIMARY_HOST, RA_PRIMARY_PORT, RA_REPLICATION_USER, RA_REPLICATION_SLOT, RA_APPLICATION_NAME. PGPASSWORD is exported only in the container authentication subprocess. The rootless wrapper sets HOME and XDG_RUNTIME_DIR for that subprocess.

Actions

Dispatcher: start, stop, monitor (status alias), promote, demote, notify, rejoin, meta-data, validate-all. Start/promote/demote/rejoin call validate_all first. Metadata advertises eight actions and omits rejoin/status; preserve this exactly.

Module ownership and dependencies

All modules are sourced once by the entrypoint before dispatch. common initializes original defaults. Source order: common, container, postgres, auth, replication, notify, promote, demote, rejoin. Functions resolve dependencies when invoked; common validation can therefore call container functions without a source cycle.

Initial extraction retains original function names and bodies to enable exact comparison, including existing unused service_home. Namespacing and recovery deduplication are deferred from this first mechanical extraction so it can serve as a verified modular baseline. No shell option, SQL, quoting, log, or error-handling cleanup is included.

common.sh

meta_data → external commands / OCF library only
validate_all → service_uid, as_service_user

container.sh

service_uid → external commands / OCF library only
service_home → external commands / OCF library only
as_service_user → external commands / OCF library only
service_running → as_service_user
container_running → as_service_user

postgres.sh

pg_command → as_service_user
pgdata_path → pg_command
pg_running → pg_command
pg_ready → as_service_user
pg_role → as_service_user
pg_wal_receiver_status → as_service_user
managed_monitor → service_running, container_running, pg_running, pg_ready, pg_role
managed_start → as_service_user, service_running, container_running, pg_command, pg_running, managed_monitor
managed_stop → as_service_user, service_running, container_running, pg_command, pg_running

auth.sh

pg_replication_exec → as_service_user
ensure_rewind_privileges → pg_command

replication.sh

pg_remote_query → pg_replication_exec
ensure_local_ha_slots → pg_command
ensure_remote_slot → pg_replication_exec
candidate_caught_up → pg_command
candidate_is_standby → pg_replication_exec

notify.sh

demote_target → external commands / OCF library only
managed_notify → external commands / OCF library only

promote.sh

managed_promote → pg_command, pg_role, managed_monitor, ensure_rewind_privileges, ensure_local_ha_slots

demote.sh

node_to_addr → external commands / OCF library only
managed_demote → as_service_user, service_running, container_running, pg_command, pg_running, pg_role, pg_wal_receiver_status, managed_monitor, node_to_addr, candidate_caught_up, demote_target, ensure_remote_slot, candidate_is_standby

rejoin.sh

managed_rejoin → as_service_user, service_running, container_running, pg_command, pg_replication_exec, pgdata_path, pg_running, pg_role, pg_remote_query, pg_wal_receiver_status, ensure_remote_slot

Incremental implementation and validation

Extract verbatim functions to the modules above in a new refactored/ tree; preserve both reference files.
Keep original OCF initialization and dispatcher; add script-relative module loading with a configured-error failure if any module cannot load.
Compare every function body, original defaults and dispatcher against the reference; run bash -n on every shell file.
Compare real meta-data output and parsed XML; exercise validate-all and record environmental limitations.
Exercise isolated action scenarios using mocked external boundaries; do not operate the live cluster.
Report hashes, install commands, and the operator-run lab comparison sequence.

5.6 KiB Raw Blame History Unescape Escape