Introducing a new way to review your PRs.

Read the announcement

usememos / memos

Memos is an open-source, self-hosted note-taking app that brings your everyday thoughts together in a personal timeline. Your data stays yours, and you choose what to keep private or share.

View on GitHub

25 documents · Updated Nov 20, 2026

docs / persistence-and-data-evolution

Persistence and data evolution

Memos keeps accounts, notes, spaces, settings, and collaboration state in a durable store. The store gives the server one product-level view of this state while database drivers, caches, migrations, transactions, and seed data keep it usable across restarts, concurrent writes, deployments, and supported database choices.

Persistence boundary

The persistence boundary separates product rules from the database that holds the state. Startup creates one store, prepares its schema, and then gives the ready store to the API and background services.

Store abstraction

The store.Store is the main access point for durable application state. It groups account, memo, space, attachment, setting, identity, inbox, reaction, share, and relation operations behind one contract. Callers use product models and search rules instead of selecting a SQL dialect or managing a database connection. References: store/store.go• L12-36, store/driver.go• L8-20, store/driver.go• L21-136

The store also protects rules that span more than one record. Writes that validate what they read run in a serializable transaction, which retries after a database lock or serialization conflict. Memo changes can update content, attachments, and relations as one durable action. User changes retain an active instance administrator, and account deletion removes the account's related records as one database action. References: store/doc.go• L5-12, store/transaction.go• L13-61, store/memo.go• L108-170, store/user.go• L13-25, store/user.go• L235-251

Persistent state model

The durable model preserves the identity and history of an instance. Core records use stable internal IDs and product-facing unique IDs where other records need durable links. Optional product data uses typed payloads stored with the record, which lets the product add fields without turning every option into a separate public model. References: store/memo.go• L47-68, store/user.go• L46-62, store/instance_setting.go• L30-73

The main state groups are:

State groupProduct purpose
Users and identitiesAccounts, roles, profiles, and sign-in links
User settingsPersonal views, themes, tags, sessions, tokens, and webhooks
Spaces and membershipsShared memo areas, roles, invitations, and access boundaries
Memos and relationsNotes, visibility, comments, links, and references
Instance settingsShared access, appearance, storage, notification, AI, and limit policy
Collaboration stateReactions, inbox messages, and memo share grants
AttachmentsFile metadata and the location of file bytes

User identity values have stable uniqueness rules across database choices. Email values are normalized before storage, while usernames remain case-sensitive. The store maps database-specific duplicate errors to stable product errors for email, username, and external identity conflicts. References: store/user.go• L109-117, store/user_unique.go• L10-67, store/…/user.go• L299-307, store/…/user.go• L307-315, store/…/user_identity_test.go• L83-170, store/user_unique_test.go• L10-51

Reaction state

Reactions are durable collaboration records that connect one user to one memo and reaction type. The store treats a repeated reaction as the same record, rejects reactions for a missing or deleted memo, and keeps authorized removal tied to the actor and memo. These rules have the same product result across SQLite, MySQL, and PostgreSQL even though each driver uses its own database operation. References: store/…/reaction.go• L13-58, store/…/reaction.go• L13-53, store/…/reaction.go• L13-59, store/…/reaction_test.go• L14-143

Memo deletion and reaction creation share a consistency boundary. A reaction cannot remain attached to a memo after the memo is removed, including when the two operations overlap. References: store/…/reaction_test.go• L279-360, store/…/reaction_test.go• L362-457

User settings

User settings keep several user-owned features in one row per user and setting key. The value is a typed protocol payload encoded as JSON, so the same store contract can hold general preferences, tags, editor choices, saved memo views, refresh tokens, personal access tokens, and webhooks. An upsert replaces the full value for that key. Memo-view updates build replacement values rather than changing cached nested data in place. References: store/user_setting.go• L15-29, store/user_setting.go• L44-83, store/user_setting.go• L494-589, store/user_setting.go• L634-747, store/…/user_setting_test.go• L805-857

Saved memo views keep their order as part of the stored setting. Add, update, remove, and reorder actions serialize the complete view list, while an invalid reorder makes no write. This keeps a user's view layout as one durable value rather than a set of partly updated records. References: store/user_setting.go• L521-653, store/user_setting.go• L656-667

Token data stays durable but follows separate product rules. Refresh-token writes remove expired entries, personal-access-token writes are serialized within a server process, and an older last-used timestamp cannot replace a newer one. A token-hash lookup returns a miss when no token matches, then joins the result to its current user. References: store/user_setting.go• L129-203, store/user_setting.go• L275-383, store/…/user_setting_test.go• L110-205, store/…/user_setting_test.go• L595-659

Database drivers keep this lookup safe for stored data from older releases. A malformed setting that does not contain the requested hash does not block other users, while a malformed row that appears to contain the requested hash returns an error instead of a false miss. References: store/…/user_setting.go• L89-128, store/…/user_setting.go• L77-113, store/…/user_setting.go• L87-143, store/…/user_setting_test.go• L15-187

Instance settings

Instance settings store shared policy as typed values in a common settings area. The server converts those values into effective settings for access, storage, notifications, AI, memo behavior, tags, and usage limits. Missing values can use safe product defaults, while storage values are normalized before the server uses them. References: store/instance_setting.go• L27-65, store/instance_setting.go• L183-218, store/instance_setting.go• L299-329

Access policy and usage limits read the durable row directly when replicas must see a change without waiting for a local cache. An absent access setting defaults to private. An absent usage-limit setting provides a 32 MiB per-file limit and an 8 KiB per-memo content limit. A stored usage-limit setting must contain a final all-users entry, so an explicit empty list is rejected rather than treated as a valid empty policy. References: store/instance_setting.go• L202-220, store/usage_limit.go• L40-132, store/…/usage_limit_test.go• L120-153

Usage limits can apply to all users or one user. The store validates the limit shape, target, values, uniqueness, and a maximum of 100 entries naming at most 1,000 users. Concurrent writers use conditional updates and retry from the latest row, so replicas do not silently replace each other's limits. Limits for deleted or missing users are removed during durable cleanup. References: store/usage_limit.go• L13-114, store/usage_limit.go• L134-186, store/…/usage_limit_test.go• L156-253

Configuration state

Declarative configuration stores its apply state in a separate system-setting row. The state has a serial and managed-resource metadata, but it is not treated as an ordinary instance setting. Before any apply, the state is empty with serial 0. References: store/configuration_state.go• L15-32, store/…/configuration_state_test.go• L26-35, store/…/configuration_state_test.go• L79-98

A configuration transaction reads its state and all related changes through one serializable transaction. It can change identity providers, instance settings, and users together, then records the next serial only after those changes pass validation. Concurrent applies retry from the latest state, and a failed apply rolls back its other writes. References: store/configuration_state.go• L34-86, store/configuration_state.go• L89-205, store/configuration_state.go• L207-279, store/…/configuration_state_test.go• L100-220

Authentication changes use the same consistency boundary. Disabling password access requires at least one identity provider, so a configuration change cannot lock regular users out. The check reads the general setting and provider list together inside the transaction. References: store/auth_config.go• L14-24, store/auth_config.go• L30-109, store/…/auth_config.go• L24-71, store/…/auth_config.go• L22-69, store/…/auth_config.go• L22-69

Cached read state

Short-lived in-process caches reduce repeated reads for users, user settings, and instance settings. The database remains the durable source. Entries use time limits and bounded item counts, and concurrent replacement, expiry, and eviction keep fresh values from being removed or counted twice. Entries leave the cache when they are replaced or deleted, and a failed write does not publish a partial value. An eviction callback can run during deletion or cleanup, and cache shutdown waits for cleanup and callback work to finish; the callback must not call shutdown synchronously. References: store/…/cache.go• L10-32, store/…/cache.go• L66-131, store/…/cache.go• L133-217, store/…/cache.go• L220-345, store/…/cache_test.go• L210-399, store/…/cache_test.go• L401-544, store/instance_setting.go• L67-99

Cached protocol values are copied before callers change nested data. Storage-setting changes also reset the derived storage driver, so a new storage choice or credential set takes effect without retaining the old driver. Access policy and usage limits bypass the ordinary setting cache when cross-replica visibility matters. References: store/user_setting.go• L494-589, store/instance_setting.go• L204-220, store/instance_setting.go• L319-329, store/storage.go• L267-327, store/storage_test.go• L317-351

Database drivers

The database layer exposes one store.Driver contract and maps it to the selected database. Each driver keeps the same product state and rules while handling its own SQL types, parameters, time values, transactions, locks, and connection behavior. References: store/driver.go• L8-20, store/…/db.go• L13-31, store/…/executor.go• L8-14, store/…/executor.go• L8-14

Driver support

Memos supports these durable database choices:

DriverDurable database roleImportant boundary
SQLiteA local database for a self-contained deploymentThe data source must be configured before opening it
MySQLA server database for shared or managed deploymentsMigration SQL must support the driver's statement rules
PostgreSQLA server database for shared or managed deploymentsThe driver uses PostgreSQL parameter and JSON behavior

MariaDB is tested through the MySQL driver and migration path. The supported compatibility target is MariaDB 11.4.13 with InnoDB and utf8mb4; conversion or replication between MySQL and MariaDB is outside this coverage. References: store/…/README.md• L1-17, store/…/containers.go• L93-165, store/…/main_test.go• L24-64

All three drivers expose the same store operations. The current schema and the incremental migration set stay aligned across them, while indexes can differ when a database type requires it. Creator indexes support account memo counts and attachment byte totals. The attachment memo lookup uses (memo_id, type) on SQLite and MySQL, and memo_id on PostgreSQL because PostgreSQL stores attachment types as unbounded text. References: store/…/memo.go• L278-286, store/…/memo.go• L308-316, store/…/memo.go• L287-295, store/…/00__add_attachment_memo_index.sql• L1, store/…/00__add_attachment_memo_index.sql• L1, store/…/00__add_attachment_memo_index.sql• L1-2, store/…/migrator_attachment_index_test.go• L16-21

The drivers also provide atomic setting primitives for first-writer initialization and conditional replacement. These primitives support safe instance-setting creation and cross-replica usage-limit and configuration-state updates without exposing database-specific conflict rules to callers. Creating a user with an external identity uses the same transaction in each driver, so the account and its identity link commit as one unit even though each database returns new values in its own way. References: store/driver.go• L60-76, store/driver.go• L131-140, store/…/instance_setting.go• L13-78, store/…/instance_setting.go• L13-87, store/…/instance_setting.go• L13-78, store/…/user_identity.go• L64-85, store/…/user_identity.go• L59-109, store/…/user_identity.go• L64-85

Database connection lifecycle

Startup validates the profile, opens the selected database, creates the common store, and migrates it before the server accepts requests. Shutdown stops cache work and closes the database connection. References: store/…/store.go• L21-37, store/store.go• L38-75, store/…/sqlite.go• L71-77

Attachment data placement

An attachment record stays in the database because it carries ownership, memo linkage, name, type, size, and storage metadata. Its bytes may be in the database or in local or S3-compatible storage, so a deployment must preserve both the database and the selected file or object store. Continue to Attachment and object storage for the binary storage lifecycle. References: store/attachment.go• L16-45, store/storage.go• L210-249

Attachment access follows the linked memo's audience and current membership. A transport-facing write checks the actor and current memo binding inside the same transaction as the metadata change. Database deletion commits the record removal first, then the store handles external bytes and derived file caches at the storage boundary. References: store/attachment.go• L47-78, store/…/attachment.go• L67-155, store/…/attachment.go• L76-164, store/…/attachment.go• L69-162

Schema evolution

Schema evolution lets a newer server change durable structure without asking each deployment to rebuild its database. A fresh database receives the current complete schema. An existing database receives only the ordered changes after its recorded schema version.

Migration baseline

The fixed migration baseline is schema 0.31.8, shipped in application release v0.31.0. It is a compatibility floor and does not advance with each application release. A database below the floor must pass through the supported intermediate releases before it can use the current server. References: store/migrator.go• L59-66, store/migrator.go• L471-505, store/…/migrator_test.go• L258-271, store/…/README.md• L1-17

Migration versioning

Incremental migrations use a calendar directory and an ordered file name. A file under YY.MM/NN__description.sql records schema version YY.M.(NN+1). The migrator validates names, sorts numeric versions, rejects duplicates, and requires every driver to ship the same version set. References: store/migrator.go• L37-43, store/migrator.go• L374-435, store/…/README.md• L19-29

The 26.11 change adds the attachment memo lookup index to existing databases. The matching current schemas keep fresh databases at the same shape. SQLite and MySQL use the memo-and-type lookup key, while PostgreSQL uses the memo key because its attachment type is unbounded text. PostgreSQL skips an existing index with the expected name, while SQLite and MySQL use ordinary index creation. References: store/…/00__add_attachment_memo_index.sql• L1, store/…/00__add_attachment_memo_index.sql• L1, store/…/00__add_attachment_memo_index.sql• L1-2, store/…/migrator_attachment_index_test.go• L23-57

Fresh database initialization

For an uninitialized database, the migrator applies that driver's LATEST.sql in a transaction and records the resulting schema version in the basic instance setting. It then initializes the access setting once. The first successful writer preserves the instance's earlier public or private behavior. References: store/migrator.go• L236-281, store/migrator.go• L142-161, store/…/migrator_test.go• L37-86, store/…/migrator_test.go• L227-295

Existing database upgrades

For an existing database, the migrator applies only files newer than the stored version and no newer than the current target. It records the new version with the migration work, so a failed upgrade does not claim success where the database supports that transaction boundary. MySQL commits DDL implicitly, so an interrupted migration can leave an index behind an old version marker and requires inspection before retry. References: store/migrator.go• L164-233, store/…/migrator_test.go• L149-225, store/…/README.md• L31-47

Upgrade checks cover baseline and calendar-release histories, preserved users, memos, settings, writable post-upgrade state, idempotent restart, and the attachment lookup indexes across SQLite, MySQL, and PostgreSQL. Application release numbers do not become schema versions. References: store/…/migrator_upgrade_test.go• L17-80, store/…/migrator_upgrade_test.go• L82-154, store/…/migrator_attachment_index_test.go• L23-91, store/…/README.md• L19-35

Upgrade compatibility

Schema version is stored in the basic instance setting. A downgrade is rejected because an older server may not understand the newer durable shape. Invalid or missing markers are also rejected, and the rejection leaves existing data and settings unchanged. References: store/migrator.go• L105-128, store/migrator.go• L447-505, store/…/migrator_guardrail_test.go• L33-85

Seed data

Seed data creates a controlled starting state for demo mode. It is separate from schema creation and runs after migration, so it can show product behavior without becoming part of the normal production data path. References: store/migrator.go• L123-139, store/migrator.go• L285-320

Demo state

Demo seeding is limited to SQLite. The seed creates representative users, Spaces, public and protected memos, comments, references, reactions, settings, and a database-stored attachment. Seed files run in name order inside a transaction. Other database drivers skip the seed path and keep their production data unchanged. References: store/migrator.go• L285-320, store/demo_seed_test.go• L37-123

The seed owns demo content rather than sign-in policy or instance settings. Declarative configuration runs first, so an existing declared administrator keeps its credentials and profile values while the seed fills only empty profile fields and attaches demo content to that user. The seed does not create user settings or access tokens. References: store/…/01__dump.sql• L1-22, store/demo_seed_test.go• L114-170

Deployment lifecycle

The deployment lifecycle joins connection setup, durable schema state, configuration state, and server startup. The order protects state on both a new deployment and an upgrade.

This diagram shows how a deployment moves from its data source to a ready server.

flowchart TD
    A[Profile and data source] --> B[Selected database driver]
    B --> C[Common store]
    C --> D{Database initialized?}
    D -- No --> E[Apply current schema]
    D -- Yes --> F[Read stored schema version]
    F --> G{Supported upgrade path?}
    G -- No --> H[Stop and use an intermediate release]
    G -- Yes --> I[Apply required migrations]
    E --> J[Record schema version]
    I --> J
    J --> K[Seed demo data when enabled]
    K --> L[Initialize access setting once]
    L --> M[Start server services]

Configuration state records the result of declarative applies without becoming a normal instance setting. It lets concurrent startup or configuration work retry from the latest serial while the transaction invalidates cached values after commit. For binary storage identities and object-store backup boundaries, continue to Attachment and object storage. For the operator upgrade path, continue to Database provisioning and migrations. Declarative configuration continues in Declarative configuration management. References: store/configuration_state.go• L15-37, store/configuration_state.go• L207-279, store/migrator.go• L96-161

Change checklist

When these files or sections are changed, remember to consider these:

  • A durable model must stay aligned across the common store model, all supported drivers, each current schema, and the incremental migration path.
  • A setting payload change must remain readable for old stored user-setting and instance-setting JSON, including token, saved-view, usage-limit, storage, and configuration-state values.
  • User-setting changes must preserve the one-row-per-user-and-key contract, full-value replacement behavior, cache invalidation, safe personal-access-token lookup, and memo-view copy and reorder behavior across all drivers.
  • Cache changes must preserve exact size accounting during concurrent replacement and expiry, protect fresh entries from stale removals, and keep eviction callbacks and shutdown ordering safe.
  • A new cross-replica setting or configuration-state update must use the driver's atomic create or conditional update contract rather than a local cache as its source of truth.
  • Usage-limit writes must preserve the 100-entry cap, 1,000-user cap, final all-users fallback, target uniqueness, cross-replica conditional updates, and cleanup for deleted users.
  • Configuration transactions must preserve transaction-local reads, shared validation, serial advancement, rollback, and post-commit invalidation for users, settings, identity providers, and storage drivers.
  • User mutations must retain an active instance administrator and keep account deletion, related rows, caches, usage limits, and external attachment cleanup consistent.
  • Memo, attachment, and reaction writes must recheck actor, ownership, membership, audience, and current bindings in one transaction where the rule spans records.
  • A migration must preserve the fixed 0.31.8 baseline, the shared version set, fresh-install schema, one-time access initialization, and supported intermediate upgrade path.
  • A migration that adds or changes a query index must appear in the current schema and in the matching incremental file for SQLite, MySQL, and PostgreSQL, with equivalent lookup behavior and driver-specific index limits respected.
  • Seed changes must remain valid against the current SQLite schema and must not become a production path for MySQL, MariaDB, or PostgreSQL.
  • MySQL-compatible changes must preserve the shared MySQL/MariaDB SQL path and the tested strict grouping behavior.

Important failure modes

TriggerConsequenceGuard
The profile selects an unknown driver or has no usable data sourceThe server cannot create the storeProfile validation and driver selection reject unsupported settings
The database has no valid marker or is below 0.31.8Startup stops and requests an intermediate upgradeThe migrator enforces the fixed baseline and old-release bridge
The database is newer than the running serverStartup rejects the downgradeThe migrator compares stored and available schema versions
Migration files have invalid names, duplicates, or driver-only versionsThe upgrade plan is ambiguous or incompleteFile validation and cross-driver version checks run before SQL
A migration statement failsThe upgrade does not advance its recorded version where transactional DDL appliesMigration work and version recording use the available transaction boundary; MySQL requires manual inspection after implicit DDL commit
An attachment index build blocks active writes during startupThe server remains unavailable until the maintenance work finishesStop other writers and run one migration instance during the maintenance window
PostgreSQL receives an attachment type too long for a type-column indexThe shared (memo_id, type) index cannot be createdPostgreSQL indexes only the bounded memo_id key
A stored user-setting or instance-setting payload cannot be decodedThe affected read fails instead of returning a partly interpreted valueTyped JSON conversion rejects invalid payloads; token lookup limits decode errors to rows that can contain the requested hash
A stored usage-limit setting is empty or has no final all-users entryThe setting is rejected instead of silently removing all limitsUsage-limit validation requires a non-empty ordered list with one final all-users entry
Two replicas update usage limits or configuration state at the same timeOne update could overwrite anotherSerializable retries and conditional writes retry against the current row
An eviction callback calls cache shutdown synchronouslyShutdown waits for the callback that is waiting for shutdownEviction callbacks request shutdown asynchronously and return first
An expired read races with a fresh replacement or in-flight cache writeA fresh value could be removed or live entries could be evictedEntry identity checks protect replacement and eviction decisions use published entries
Password sign-in is disabled with no effective identity providerRegular users can be locked outStored authentication state and identity providers are checked together
A creator- or memo-based query lacks its driver indexCounts, quota checks, or attachment filters can become slow on large data setsCurrent schemas and the 26.10 and 26.11 migrations add the required lookup indexes
A reaction is created while its memo is being removedCollaboration state could point to a removed memoDriver upserts check the memo inside the write boundary and memo cleanup removes reactions
A caller reads an attachment outside the memo audiencePrivate file metadata or bytes can be exposedDriver queries apply memo access rules before pagination
A user deletion fails after related rows are preparedThe account and its related state must remain intactRelated deletion runs in one transaction and cache cleanup follows success
Demo mode uses MySQL, MariaDB, or PostgreSQLProduction data could be mistaken for demo dataSeed application is limited to SQLite
A duplicate user or identity value reaches a database driverCallers receive a raw, driver-specific uniqueness errorclassifyUserUniqueViolation maps supported SQLite, MySQL, MariaDB, and PostgreSQL markers to stable store errors

Key files

FileWhy to read it
store/doc.goDefines the persistence layer boundary and transaction ownership
store/store.goShows the durable store boundary, caches, storage-driver reuse, and shutdown behavior
store/driver.goDefines the database-independent contract and atomic setting primitives
store/transaction.goShows serializable transaction retries
store/configuration_state.goShows declarative state, serials, transaction-local reads, and post-commit invalidation
store/auth_config.goShows safe authentication configuration mutations
store/memo.goDefines memo state, visibility, access, and write policy
store/attachment.goDefines attachment metadata and the database-to-file boundary
store/storage.goShows storage identities, legacy compatibility, and storage-driver cache behavior
store/user.goShows user normalization, admin retention, deletion, and limit cleanup
store/user_unique.goMaps database-specific user and identity conflicts to store errors
store/user_setting.goShows typed user settings, token state, saved views, memo-view updates, and cache updates
store/usage_limit.goShows instance and per-user usage-limit state and concurrent writes
store/instance_setting.goShows typed settings, defaults, cache use, and storage-driver reset
store/…/cache.goDefines the cache contract, TTL, bounded size, concurrency, eviction, and shutdown behavior
store/…/cache_test.goShows cache expiry, bounds, concurrent accounting, eviction, and shutdown guarantees
store/…/memo.goShows MySQL memo persistence, filtering, payload handling, and memo deletion cleanup
store/…/space.goShows MySQL space membership and relationship writes with transaction boundaries
store/…/reaction.goShows SQLite reaction uniqueness and memo-lifecycle behavior
store/…/reaction.goShows MySQL reaction writes and transaction boundaries
store/…/reaction.goShows PostgreSQL reaction writes and transaction boundaries
store/…/instance_setting.goShows SQLite setting storage and conditional writes
store/…/instance_setting.goShows MySQL setting storage and exact value comparison
store/…/instance_setting.goShows PostgreSQL setting storage and row locks
store/…/auth_config.goShows SQLite authentication mutation and retry behavior
store/…/auth_config.goShows MySQL authentication mutation and retry behavior
store/…/auth_config.goShows PostgreSQL authentication mutation and retry behavior
store/…/user_identity.goShows SQLite user and external identity creation as one transaction
store/…/user_identity.goShows MySQL user and external identity creation with database-specific value read-back
store/…/user_identity.goShows PostgreSQL user and external identity creation with returned values
store/migrator.goExplains initialization, upgrades, version tracking, and seeding
store/…/README.mdDefines the baseline, migration naming contract, and upgrade maintenance rules
store/…/00__add_attachment_memo_index.sqlShows the SQLite attachment lookup index
store/…/00__add_attachment_memo_index.sqlShows the MySQL attachment lookup index
store/…/00__add_attachment_memo_index.sqlShows the PostgreSQL index shape for unbounded attachment types
store/…/store.goShows isolated test databases and driver-specific store setup
store/…/containers.goShows MySQL, MariaDB, PostgreSQL, and released-binary test environments
store/…/main_test.goShows the cross-driver test matrix
store/…/migrator_test.goShows fresh-install, repeat-run, data-preservation, and initialization guarantees
store/…/migrator_attachment_index_test.goShows index upgrade, restart, and attachment-data preservation guarantees
store/…/migrator_guardrail_test.goShows rejection of unsupported schema histories without data changes
store/…/migrator_upgrade_test.goShows upgrade and restart preservation across release histories
store/…/memo_filter_test.goShows the durable memo fields and payload values that the filter surface must preserve across drivers
store/…/user_identity_test.goShows identity uniqueness and atomic user-plus-identity creation
store/user_unique_test.goShows stable conflict mapping across database error formats
store/…/README.mdDefines the supported store test targets and upgrade coverage
store/…/configuration_state_test.goShows state serials, concurrent applies, rollback, and cache visibility
store/…/instance_setting_test.goShows setting defaults, first-writer behavior, copy isolation, and storage cache reset
store/…/reaction_test.goShows reaction uniqueness, authorization, memo deletion, and concurrent cleanup guarantees
store/storage_test.goShows storage identity preservation and legacy attachment resolution
store/demo_seed_test.goShows the controlled SQLite demo state
docs/attachment-and-object-storage.mdContinues the binary attachment storage and backup model
docs/database-provisioning-and-migrations.mdContinues the operational database upgrade path
docs/declarative-configuration-management.mdContinues the declarative configuration workflow