fix for retention logic

This commit is contained in:
onysd 2026-09-03 05:00:42 +03:00
parent 70c0ba44f0
commit e6bfe2d444
35 changed files with 2108 additions and 132 deletions

View file

@ -12,7 +12,12 @@ import (
)
func (s *ChannelStore) EditChannelMessage(ctx context.Context, req domain.EditChannelMessageRequest) (domain.EditChannelMessageResult, error) {
if req.UserID == 0 || req.ChannelID == 0 || req.ID <= 0 {
retentionPurge := req.RetentionPurge && req.RetentionPurgeAction != nil
// RetentionPurge is a server-internal edit with no acting user (the
// storage retention sweep, not an RPC caller) -- UserID==0 is normally
// invalid, and req.UserID would otherwise also need to name an existing
// channel member for getChannelForMember below.
if (req.UserID == 0 && !retentionPurge) || req.ChannelID == 0 || req.ID <= 0 {
return domain.EditChannelMessageResult{}, domain.ErrChannelInvalid
}
beginner, ok := s.db.(txBeginner)
@ -44,7 +49,17 @@ func (s *ChannelStore) EditChannelMessage(ctx context.Context, req domain.EditCh
_ = tx.Rollback(ctx)
}
}()
channel, member, err := s.getChannelForMember(ctx, tx, req.UserID, req.ChannelID)
var (
channel domain.Channel
member domain.ChannelMember
)
if retentionPurge {
// No membership requirement: this is a server-internal edit, not an
// action taken by any particular channel member.
channel, err = s.channelByID(ctx, tx, req.ChannelID)
} else {
channel, member, err = s.getChannelForMember(ctx, tx, req.UserID, req.ChannelID)
}
if err != nil {
return domain.EditChannelMessageResult{}, err
}
@ -52,6 +67,17 @@ func (s *ChannelStore) EditChannelMessage(ctx context.Context, req domain.EditCh
if err != nil {
return domain.EditChannelMessageResult{}, err
}
if retentionPurge {
if msg.Deleted {
return domain.EditChannelMessageResult{}, domain.ErrMessageIDInvalid
}
if msg.Action != nil {
// 幂等守卫:已经是服务消息(含已被本路径转换过的 retention 通知)
// 不再重复编辑 -- 见 EditMessageRequest.RetentionPurge 同款守卫。
return domain.EditChannelMessageResult{}, domain.ErrMessageNotModified
}
return s.applyRetentionPurgeChannelMessage(ctx, tx, channel, msg, req)
}
if msg.Deleted || msg.Action != nil {
return domain.EditChannelMessageResult{}, domain.ErrMessageIDInvalid
}
@ -283,6 +309,74 @@ WHERE channel_id = $1 AND id = $2`, req.ChannelID, req.ID, mediaJSON); err != ni
return domain.EditChannelMessageResult{Channel: channel, Message: msg, Event: event, ServiceMessage: serviceMsg, ServiceEvent: serviceEvent, Recipients: recipients}, nil
}
// applyRetentionPurgeChannelMessage turns msg into a real service message
// (messageActionCustomAction via req.RetentionPurgeAction) in place. Unlike a
// normal channel edit -- which only ever replaces body/entities/media --
// channel service actions live in a structurally separate Action column
// (see tgChannelMessage: m.Action != nil renders tg.MessageService instead of
// tg.Message), so this clears body/media and sets Action instead of touching
// them. Reuses the same pts/durable-event/admin-log machinery as a normal
// edit so getChannelDifference and the fanout dispatcher see it identically.
func (s *ChannelStore) applyRetentionPurgeChannelMessage(ctx context.Context, tx pgx.Tx, channel domain.Channel, msg domain.ChannelMessage, req domain.EditChannelMessageRequest) (domain.EditChannelMessageResult, error) {
actionJSON, err := marshalJSON(req.RetentionPurgeAction, "{}")
if err != nil {
return domain.EditChannelMessageResult{}, fmt.Errorf("encode retention purge channel action: %w", err)
}
pts, err := s.reserveChannelPts(ctx, tx, req.ChannelID)
if err != nil {
return domain.EditChannelMessageResult{}, fmt.Errorf("allocate retention purge channel pts: %w", err)
}
if _, err := tx.Exec(ctx, `
UPDATE channel_messages
SET body = '', entities = '[]'::jsonb, media = '{}'::jsonb, action = $3, edit_date = $4, pts = $5, updated_at = now()
WHERE channel_id = $1 AND id = $2`, req.ChannelID, req.ID, actionJSON, req.EditDate, pts); err != nil {
return domain.EditChannelMessageResult{}, fmt.Errorf("update retention purge channel message: %w", err)
}
prevMsg := msg
msg.Body = ""
msg.Entities = nil
msg.Media = nil
msg.Action = req.RetentionPurgeAction
msg.EditDate = req.EditDate
msg.Pts = pts
// 媒体索引/media_references 靠这次替换后的(空)媒体重建:原文档/照片
// 不再被这条消息引用storage retention 的孤儿判定据此推进。
if err := replaceChannelMediaIndexTx(ctx, tx, req.ChannelID, req.ID, msg.Date, msg.Media, nil); err != nil {
return domain.EditChannelMessageResult{}, err
}
event := domain.ChannelUpdateEvent{
ChannelID: req.ChannelID,
Type: domain.ChannelUpdateEditMessage,
Pts: pts,
PtsCount: 1,
Date: req.EditDate,
Message: msg,
SenderUserID: req.UserID,
}
if err := insertChannelEventTx(ctx, tx, event); err != nil {
return domain.EditChannelMessageResult{}, err
}
if _, err := tx.Exec(ctx, `UPDATE channels SET pts = $2, updated_at = now() WHERE id = $1`, req.ChannelID, pts); err != nil {
return domain.EditChannelMessageResult{}, fmt.Errorf("update retention purge channel pts: %w", err)
}
channel.Pts = pts
if err := s.insertChannelAdminLogTx(ctx, tx, domain.ChannelAdminLogEvent{
ChannelID: req.ChannelID,
UserID: req.UserID,
Date: req.EditDate,
Type: domain.ChannelAdminLogEditMessage,
PrevMessage: &prevMsg,
NewMessage: &msg,
}); err != nil {
return domain.EditChannelMessageResult{}, err
}
if err := tx.Commit(ctx); err != nil {
return domain.EditChannelMessageResult{}, fmt.Errorf("commit retention purge channel message: %w", err)
}
recipients, _ := s.ListActiveChannelMemberIDs(ctx, req.UserID, req.ChannelID, 0)
return domain.EditChannelMessageResult{Channel: channel, Message: msg, Event: event, Recipients: recipients}, nil
}
func isChannelTodoParticipantEdit(req domain.EditChannelMessageRequest, msg domain.ChannelMessage) bool {
if !req.AllowTodoParticipantMutation || req.SetReplyMarkup || req.Media == nil || req.Media.Kind != domain.MessageMediaKindTodo || req.Media.Todo == nil {
return false

View file

@ -64,7 +64,7 @@ func TestHardRetentionPurgesBlobBytesButKeepsMetadataRow(t *testing.T) {
})
cutoff := time.Now().Add(-24 * time.Hour)
ids, err := media.ListDocumentIDsForHardRetentionOlderThan(ctx, cutoff, 1000)
ids, err := media.ListDocumentIDsForHardRetentionOlderThan(ctx, domain.MediaCategoryNone, cutoff, 1000)
if err != nil {
t.Fatalf("ListDocumentIDsForHardRetentionOlderThan: %v", err)
}
@ -95,7 +95,7 @@ func TestHardRetentionPurgesBlobBytesButKeepsMetadataRow(t *testing.T) {
// Idempotent / self-terminating: a second sweep pass no longer selects
// this document, since it no longer owns any file_blobs row.
ids, err = media.ListDocumentIDsForHardRetentionOlderThan(ctx, cutoff, 1000)
ids, err = media.ListDocumentIDsForHardRetentionOlderThan(ctx, domain.MediaCategoryNone, cutoff, 1000)
if err != nil {
t.Fatalf("ListDocumentIDsForHardRetentionOlderThan (2nd pass): %v", err)
}

View file

@ -24,9 +24,30 @@ ON CONFLICT (channel_id, id, category) DO NOTHING`, channelID, id, int16(c), dat
return fmt.Errorf("insert channel media index: %w", err)
}
}
if err := setDocumentCategoryTx(ctx, tx, media); err != nil {
return err
}
return addMediaReferencesTx(ctx, tx, media, domain.MediaRefKindChannelMessage, channelMessageRefKey(channelID, id))
}
// setDocumentCategoryTx persists media.Document's retention-sweep category
// (see domain.DocumentMediaCategory) onto its documents row, alongside the
// message_box_media/channel_message_media index write these callers already
// do -- zero new classification logic, just also stamping the already
// computed value directly onto documents.category so the per-category
// storage retention sweep (internal/app/files/retention.go) can filter on it
// without a join. No-op for non-document media.
func setDocumentCategoryTx(ctx context.Context, tx pgx.Tx, media *domain.MessageMedia) error {
if media == nil || media.Kind != domain.MessageMediaKindDocument || media.Document == nil || media.Document.ID == 0 {
return nil
}
category := domain.DocumentMediaCategory(media)
if _, err := tx.Exec(ctx, `UPDATE documents SET category = $2 WHERE id = $1`, media.Document.ID, int16(category)); err != nil {
return fmt.Errorf("set document category: %w", err)
}
return nil
}
// deleteChannelMediaIndexTx 清掉一条频道消息的全部索引行(编辑改媒体前先清后插)。
// 只在 replaceChannelMediaIndexTx 内被调用;真正的消息删除不清 *_media 分类索引
// (读时靠 JOIN deleted 过滤,见文件头注释),但仍需在此清掉 media_references,
@ -57,6 +78,9 @@ ON CONFLICT (owner_user_id, box_id, category) DO NOTHING`, ownerUserID, boxID, p
return fmt.Errorf("insert message box media index: %w", err)
}
}
if err := setDocumentCategoryTx(ctx, tx, media); err != nil {
return err
}
return addMediaReferencesTx(ctx, tx, media, domain.MediaRefKindMessageBox, messageBoxRefKey(ownerUserID, boxID))
}

View file

@ -105,6 +105,30 @@ RETURNING media_kind, media_id`, string(refKind), refKey)
return nil
}
// ListMediaReferences returns every live reference row for (kind, mediaID) --
// used by the storage retention sweep to turn a hard-retention/eviction blob
// purge into a visible notice on every message that still embeds the purged
// media. See internal/app/files.notifyRetentionPurge.
func (s *MediaStore) ListMediaReferences(ctx context.Context, kind domain.MediaKind, mediaID int64) ([]domain.MediaReference, error) {
rows, err := s.q.ListMediaReferences(ctx, sqlcgen.ListMediaReferencesParams{
MediaKind: string(kind),
MediaID: mediaID,
})
if err != nil {
return nil, fmt.Errorf("list media references: %w", err)
}
out := make([]domain.MediaReference, 0, len(rows))
for _, r := range rows {
out = append(out, domain.MediaReference{
Kind: domain.MediaKind(r.MediaKind),
MediaID: r.MediaID,
RefKind: domain.MediaRefKind(r.RefKind),
RefKey: r.RefKey,
})
}
return out, nil
}
// ---- storage retention sweep ----
// OrphanDocumentIfUnreferenced marks a document orphaned right now if
@ -128,20 +152,22 @@ WHERE id = $1
return tag.RowsAffected() > 0, nil
}
// ListOrphanedDocumentIDsOlderThan returns document ids whose orphaned_at is
// set and older than cutoff, oldest first, up to limit.
func (s *MediaStore) ListOrphanedDocumentIDsOlderThan(ctx context.Context, cutoff time.Time, limit int) ([]int64, error) {
// ListOrphanedDocumentIDsOlderThan returns document ids in the given category
// whose orphaned_at is set and older than cutoff, oldest first, up to limit.
func (s *MediaStore) ListOrphanedDocumentIDsOlderThan(ctx context.Context, category domain.MediaCategory, cutoff time.Time, limit int) ([]int64, error) {
if limit <= 0 {
return nil, nil
}
return s.q.ListOrphanedDocumentIDsOlderThan(ctx, sqlcgen.ListOrphanedDocumentIDsOlderThanParams{
Cutoff: pgtype.Timestamptz{Time: cutoff, Valid: true},
Category: int16(category),
BatchLimit: int32(limit),
})
}
// ListOrphanedPhotoIDsOlderThan returns photo ids whose orphaned_at is set
// and older than cutoff, oldest first, up to limit.
// ListOrphanedPhotoIDsOlderThan returns photo ids (excluding a live avatar --
// see ListAvatarOrphanedPhotoIDsOlderThan) whose orphaned_at is set and older
// than cutoff, oldest first, up to limit.
func (s *MediaStore) ListOrphanedPhotoIDsOlderThan(ctx context.Context, cutoff time.Time, limit int) ([]int64, error) {
if limit <= 0 {
return nil, nil
@ -152,6 +178,19 @@ func (s *MediaStore) ListOrphanedPhotoIDsOlderThan(ctx context.Context, cutoff t
})
}
// ListAvatarOrphanedPhotoIDsOlderThan is the "Avatar" category counterpart of
// ListOrphanedPhotoIDsOlderThan -- see the query's doc comment for why this
// is expected to stay empty in practice under "orphan" mode.
func (s *MediaStore) ListAvatarOrphanedPhotoIDsOlderThan(ctx context.Context, cutoff time.Time, limit int) ([]int64, error) {
if limit <= 0 {
return nil, nil
}
return s.q.ListAvatarOrphanedPhotoIDsOlderThan(ctx, sqlcgen.ListAvatarOrphanedPhotoIDsOlderThanParams{
Cutoff: pgtype.Timestamptz{Time: cutoff, Valid: true},
BatchLimit: int32(limit),
})
}
// CountFileBlobRefs reports how many file_blobs rows still point at
// (backend, objectKey) -- the caller must not physically delete the object
// from that backend while this is > 0 (content-addressed storage: the same
@ -197,24 +236,26 @@ func (s *MediaStore) DeleteDocumentAndBlobs(ctx context.Context, id int64) ([]do
return blobs, nil
}
// ListDocumentIDsForHardRetentionOlderThan returns document ids older than
// cutoff (by upload/created_at) that still own at least one file_blobs row,
// oldest first, up to limit -- the "hard" retention sweep's candidate list.
// Unlike ListOrphanedDocumentIDsOlderThan, this ignores media_references
// entirely: a document still referenced by a live message is exactly as
// eligible as an orphaned one.
func (s *MediaStore) ListDocumentIDsForHardRetentionOlderThan(ctx context.Context, cutoff time.Time, limit int) ([]int64, error) {
// ListDocumentIDsForHardRetentionOlderThan returns document ids in the given
// category older than cutoff (by upload/created_at) that still own at least
// one file_blobs row, oldest first, up to limit -- the "hard" retention
// sweep's candidate list. Unlike ListOrphanedDocumentIDsOlderThan, this
// ignores media_references entirely: a document still referenced by a live
// message is exactly as eligible as an orphaned one.
func (s *MediaStore) ListDocumentIDsForHardRetentionOlderThan(ctx context.Context, category domain.MediaCategory, cutoff time.Time, limit int) ([]int64, error) {
if limit <= 0 {
return nil, nil
}
return s.q.ListDocumentIDsForHardRetentionOlderThan(ctx, sqlcgen.ListDocumentIDsForHardRetentionOlderThanParams{
Cutoff: pgtype.Timestamptz{Time: cutoff, Valid: true},
Category: int16(category),
BatchLimit: int32(limit),
})
}
// ListPhotoIDsForHardRetentionOlderThan is the photo counterpart of
// ListDocumentIDsForHardRetentionOlderThan.
// ListDocumentIDsForHardRetentionOlderThan (excluding a live avatar -- see
// ListAvatarPhotoIDsForHardRetentionOlderThan).
func (s *MediaStore) ListPhotoIDsForHardRetentionOlderThan(ctx context.Context, cutoff time.Time, limit int) ([]int64, error) {
if limit <= 0 {
return nil, nil
@ -225,6 +266,18 @@ func (s *MediaStore) ListPhotoIDsForHardRetentionOlderThan(ctx context.Context,
})
}
// ListAvatarPhotoIDsForHardRetentionOlderThan is the "Avatar" category
// counterpart of ListPhotoIDsForHardRetentionOlderThan.
func (s *MediaStore) ListAvatarPhotoIDsForHardRetentionOlderThan(ctx context.Context, cutoff time.Time, limit int) ([]int64, error) {
if limit <= 0 {
return nil, nil
}
return s.q.ListAvatarPhotoIDsForHardRetentionOlderThan(ctx, sqlcgen.ListAvatarPhotoIDsForHardRetentionOlderThanParams{
Cutoff: pgtype.Timestamptz{Time: cutoff, Valid: true},
BatchLimit: int32(limit),
})
}
// DeleteFileBlobsForDocument deletes every file_blobs row a document owns
// (main body + thumbnail variants), returning what was deleted so the caller
// can physically remove each object from its backend once confirming (via
@ -292,6 +345,32 @@ func (s *MediaStore) DeleteFileBlobsForPhoto(ctx context.Context, id int64) ([]d
return blobs, nil
}
// ListOldestMediaForEviction returns up to limit documents and limit photos
// still owning file_blobs bytes, oldest-uploaded first, for the active
// eviction sweep to interleave by actual created_at (not drain one table
// before the other) -- see domain.EvictionCandidate.
func (s *MediaStore) ListOldestMediaForEviction(ctx context.Context, limit int) ([]domain.EvictionCandidate, error) {
if limit <= 0 {
return nil, nil
}
docs, err := s.q.ListOldestDocumentsForEviction(ctx, int32(limit))
if err != nil {
return nil, fmt.Errorf("list oldest documents for eviction: %w", err)
}
photos, err := s.q.ListOldestPhotosForEviction(ctx, int32(limit))
if err != nil {
return nil, fmt.Errorf("list oldest photos for eviction: %w", err)
}
out := make([]domain.EvictionCandidate, 0, len(docs)+len(photos))
for _, d := range docs {
out = append(out, domain.EvictionCandidate{Kind: domain.MediaKindDocument, MediaID: d.ID, CreatedAt: d.CreatedAt.Time})
}
for _, p := range photos {
out = append(out, domain.EvictionCandidate{Kind: domain.MediaKindPhoto, MediaID: p.ID, CreatedAt: p.CreatedAt.Time})
}
return out, nil
}
// DeletePhotoAndBlobs deletes a photo row and every file_blobs row it owns
// (one per rendition size), returning what was deleted so the caller can
// physically remove each object from its backend once confirming (via

View file

@ -83,9 +83,18 @@ func (s *MessageStore) EditMessage(ctx context.Context, req domain.EditMessageRe
}
authorEdit := target.Outgoing && target.MessageSenderID == req.OwnerUserID && target.FromUserID == req.OwnerUserID
viaBotEdit := req.ViaBotEditBotID != 0 && target.ViaBotID == req.ViaBotEditBotID
if !authorEdit && !viaBotEdit && !req.WebPageResolve && !validTodoParticipantEdit(req, target, oldEntities) {
if !authorEdit && !viaBotEdit && !req.WebPageResolve && !req.RetentionPurge && !validTodoParticipantEdit(req, target, oldEntities) {
return res, domain.ErrMessageAuthorRequired
}
if req.RetentionPurge {
// 幂等守卫:已经是这条 retention 通知的消息不再重复编辑(同一被回收
// 媒体可能被多个 box 的 media_references 行各引用一次,比如自己与对端
// 各自的 box——同一条共享 private_message 只需真正编辑一次)。
if targetMedia, err := decodeMessageMedia(target.MediaJson); err == nil &&
targetMedia != nil && targetMedia.Kind == domain.MessageMediaKindService {
return res, domain.ErrMessageNotModified
}
}
richChanged := req.SetRichMessage && !richMessagesEqual(targetRich, req.RichMessage)
if req.Media == nil && !req.SetReplyMarkup && !richChanged && target.Body == req.Message && target.HideEdited == req.HideEdited && sameMessageEntities(oldEntities, req.Entities) {
return res, domain.ErrMessageNotModified

View file

@ -202,6 +202,16 @@ WHERE id = sqlc.arg(media_id)::bigint AND orphaned_at IS NOT NULL;
UPDATE photos SET orphaned_at = NULL
WHERE id = sqlc.arg(media_id)::bigint AND orphaned_at IS NOT NULL;
-- name: ListMediaReferences :many
-- Every live reference to a document/photo -- used by the storage retention
-- sweep to turn a hard-retention/eviction blob purge into a visible notice on
-- every message that still embeds the purged media (see
-- files.notifyRetentionPurge). profile_photo/sticker_set/gift refs have no
-- message to edit and are filtered by the caller, not this query.
SELECT media_kind, media_id, ref_kind, ref_key
FROM media_references
WHERE media_kind = sqlc.arg(media_kind)::text AND media_id = sqlc.arg(media_id)::bigint;
-- name: RemoveMediaReference :exec
DELETE FROM media_references
WHERE media_kind = sqlc.arg(media_kind)::text
@ -226,15 +236,36 @@ WHERE id = sqlc.arg(media_id)::bigint
);
-- name: ListOrphanedDocumentIDsOlderThan :many
-- category selects one documents.category bucket per sweep tick (per-category
-- retention age, see internal/app/files/retention.go) -- 0 (MediaCategoryNone)
-- covers unclassified documents (e.g. stickers), which always use the shared
-- global age since there is no per-category override for that bucket.
SELECT id FROM documents
WHERE orphaned_at IS NOT NULL AND orphaned_at < sqlc.arg(cutoff)::timestamptz
AND category = sqlc.arg(category)::smallint
ORDER BY orphaned_at ASC
LIMIT sqlc.arg(batch_limit)::int;
-- name: ListOrphanedPhotoIDsOlderThan :many
SELECT id FROM photos
WHERE orphaned_at IS NOT NULL AND orphaned_at < sqlc.arg(cutoff)::timestamptz
ORDER BY orphaned_at ASC
-- Excludes photos currently active as someone's avatar -- see
-- ListAvatarOrphanedPhotoIDsOlderThan for that split-off bucket.
SELECT p.id FROM photos p
WHERE p.orphaned_at IS NOT NULL AND p.orphaned_at < sqlc.arg(cutoff)::timestamptz
AND NOT EXISTS (SELECT 1 FROM profile_photos pp WHERE pp.photo_id = p.id AND pp.active)
ORDER BY p.orphaned_at ASC
LIMIT sqlc.arg(batch_limit)::int;
-- name: ListAvatarOrphanedPhotoIDsOlderThan :many
-- Same as ListOrphanedPhotoIDsOlderThan but only photos currently active as
-- someone's avatar (profile_photos.active) -- lets the Avatar category carry
-- its own retention age. In practice a live avatar is never orphaned (an
-- active profile_photos row is itself a media_references entry), so this
-- bucket is expected to stay empty under "orphan" mode; kept for symmetry
-- with the "hard" mode avatar split, which is the one that actually matters.
SELECT p.id FROM photos p
WHERE p.orphaned_at IS NOT NULL AND p.orphaned_at < sqlc.arg(cutoff)::timestamptz
AND EXISTS (SELECT 1 FROM profile_photos pp WHERE pp.photo_id = p.id AND pp.active)
ORDER BY p.orphaned_at ASC
LIMIT sqlc.arg(batch_limit)::int;
-- name: ListDocumentIDsForHardRetentionOlderThan :many
@ -244,9 +275,12 @@ LIMIT sqlc.arg(batch_limit)::int;
-- keeps this sweep from re-selecting the same document forever: once its
-- blob bytes are purged (DeleteFileBlobsForDocument removes the file_blobs
-- rows but deliberately leaves this documents row in place), it naturally
-- drops out of this query on the next pass.
-- drops out of this query on the next pass. category scopes to one
-- documents.category bucket per sweep tick -- see
-- ListOrphanedDocumentIDsOlderThan for the 0/None fallback note.
SELECT d.id FROM documents d
WHERE d.created_at < sqlc.arg(cutoff)::timestamptz
AND d.category = sqlc.arg(category)::smallint
AND EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'doc:' || d.id::text
@ -257,9 +291,11 @@ LIMIT sqlc.arg(batch_limit)::int;
-- name: ListPhotoIDsForHardRetentionOlderThan :many
-- See ListDocumentIDsForHardRetentionOlderThan -- same "hard" retention
-- candidate selection, for photos.
-- candidate selection, for photos. Excludes photos currently active as
-- someone's avatar -- see ListAvatarPhotoIDsForHardRetentionOlderThan.
SELECT p.id FROM photos p
WHERE p.created_at < sqlc.arg(cutoff)::timestamptz
AND NOT EXISTS (SELECT 1 FROM profile_photos pp WHERE pp.photo_id = p.id AND pp.active)
AND EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'photo:' || p.id::text
@ -268,6 +304,52 @@ WHERE p.created_at < sqlc.arg(cutoff)::timestamptz
ORDER BY p.created_at ASC
LIMIT sqlc.arg(batch_limit)::int;
-- name: ListAvatarPhotoIDsForHardRetentionOlderThan :many
-- Same as ListPhotoIDsForHardRetentionOlderThan but only photos currently
-- active as someone's avatar (profile_photos.active) -- lets the Avatar
-- category carry its own retention age, independent of ordinary shared-media
-- photos.
SELECT p.id FROM photos p
WHERE p.created_at < sqlc.arg(cutoff)::timestamptz
AND EXISTS (SELECT 1 FROM profile_photos pp WHERE pp.photo_id = p.id AND pp.active)
AND EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'photo:' || p.id::text
OR fb.location_key LIKE 'photo:' || p.id::text || ':%'
)
ORDER BY p.created_at ASC
LIMIT sqlc.arg(batch_limit)::int;
-- name: ListOldestDocumentsForEviction :many
-- Active eviction (TELESRV_STORAGE_EVICTION_ENABLE): candidates are every
-- document that still owns at least one file_blobs row, oldest-uploaded
-- first, REGARDLESS of category or age -- unlike the retention sweeps above,
-- eviction only cares about reclaiming bytes once the total physical budget
-- (TELESRV_STORAGE_MAX_TOTAL_BYTES) is exceeded. created_at is returned so
-- the caller can interleave these with ListOldestPhotosForEviction by actual
-- age instead of draining one table before touching the other.
SELECT d.id, d.created_at FROM documents d
WHERE EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'doc:' || d.id::text
OR fb.location_key LIKE 'doc:' || d.id::text || ':%'
)
ORDER BY d.created_at ASC
LIMIT sqlc.arg(batch_limit)::int;
-- name: ListOldestPhotosForEviction :many
-- See ListOldestDocumentsForEviction -- same oldest-first eviction candidate
-- selection, for photos (avatars included: eviction is bytes-only and does
-- not honor the Avatar category's separate retention age).
SELECT p.id, p.created_at FROM photos p
WHERE EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'photo:' || p.id::text
OR fb.location_key LIKE 'photo:' || p.id::text || ':%'
)
ORDER BY p.created_at ASC
LIMIT sqlc.arg(batch_limit)::int;
-- name: CountFileBlobRefs :one
SELECT COUNT(*)::int FROM file_blobs WHERE backend = sqlc.arg(backend)::text AND object_key = sqlc.arg(object_key)::text;

View file

@ -0,0 +1,360 @@
package postgres
import (
"context"
"strconv"
"testing"
"time"
"telesrv/internal/domain"
)
// TestRetentionPurgeBypassProducesServiceMessage_PrivateMessage exercises
// Part 1 of the storage retention v2 plan end-to-end against a real
// Postgres: EditMessage with RetentionPurge=true bypasses the ordinary
// author check (the storage retention sweep is not the message's author) and
// replaces the message media with a messageActionCustomAction service
// payload, and a second retention-purge edit on the now-already-service
// message is a no-op (ErrMessageNotModified) -- guarding the idempotency
// guard that keeps a shared media_references duplicate (owner+peer both
// referencing the same purged document) from double-editing the same box.
func TestRetentionPurgeBypassProducesServiceMessage_PrivateMessage(t *testing.T) {
pool := testPool(t)
ctx := context.Background()
suffix := randomSuffix(t)
users := NewUserStore(pool)
sender, err := users.Create(ctx, domain.User{AccessHash: 81, Phone: "+1997" + suffix + "01", FirstName: "RetentionSender"})
if err != nil {
t.Fatalf("create sender: %v", err)
}
recipient, err := users.Create(ctx, domain.User{AccessHash: 82, Phone: "+1997" + suffix + "02", FirstName: "RetentionRecipient"})
if err != nil {
t.Fatalf("create recipient: %v", err)
}
ids := []int64{sender.ID, recipient.ID}
t.Cleanup(func() {
_, _ = pool.Exec(ctx, "DELETE FROM message_boxes WHERE owner_user_id = ANY($1::bigint[])", ids)
_, _ = pool.Exec(ctx, "DELETE FROM private_messages WHERE sender_user_id = ANY($1::bigint[])", ids)
_, _ = pool.Exec(ctx, "DELETE FROM dialogs WHERE user_id = ANY($1::bigint[])", ids)
_, _ = pool.Exec(ctx, "DELETE FROM users WHERE id = ANY($1::bigint[])", ids)
})
messages := NewMessageStore(pool, WithMessageAllocators(&perUserCounterAllocator{}))
sent, err := messages.SendPrivateText(ctx, domain.SendPrivateTextRequest{
SenderUserID: sender.ID,
RecipientUserID: recipient.ID,
RandomID: time.Now().UnixNano(),
Message: "a file worth keeping",
Media: &domain.MessageMedia{
Kind: domain.MessageMediaKindDocument,
Document: &domain.Document{ID: time.Now().UnixNano(), AccessHash: 1, MimeType: "application/octet-stream"},
},
Date: int(time.Now().Unix()),
})
if err != nil {
t.Fatalf("send private media: %v", err)
}
noticeMedia := &domain.MessageMedia{
Kind: domain.MessageMediaKindService,
ServiceAction: &domain.MessageServiceAction{
Kind: domain.MessageServiceActionCustomText,
Text: domain.RetentionPurgeNoticeText,
},
}
req := domain.EditMessageRequest{
OwnerUserID: sender.ID,
Peer: domain.Peer{Type: domain.PeerTypeUser, ID: recipient.ID},
ID: sent.SenderMessage.ID,
SetRichMessage: true,
Media: noticeMedia,
RetentionPurge: true,
}
res, err := messages.EditMessage(ctx, req)
if err != nil {
t.Fatalf("EditMessage(RetentionPurge): %v", err)
}
self := res.Self()
if self.Message.Media == nil || self.Message.Media.Kind != domain.MessageMediaKindService {
t.Fatalf("edited message media = %+v, want service kind", self.Message.Media)
}
if self.Message.Media.ServiceAction == nil || self.Message.Media.ServiceAction.Kind != domain.MessageServiceActionCustomText {
t.Fatalf("edited message service action = %+v, want custom_text", self.Message.Media.ServiceAction)
}
if self.Message.Media.ServiceAction.Text != domain.RetentionPurgeNoticeText {
t.Fatalf("edited message notice text = %q, want %q", self.Message.Media.ServiceAction.Text, domain.RetentionPurgeNoticeText)
}
// A second retention-purge edit is a no-op: the message is already the
// notice, so this must not keep bumping pts / re-writing it forever.
if _, err := messages.EditMessage(ctx, req); err != domain.ErrMessageNotModified {
t.Fatalf("second RetentionPurge edit err = %v, want ErrMessageNotModified", err)
}
// A non-author, non-bypass edit on someone else's message must still be
// rejected -- RetentionPurge only bypasses the author check when it is
// actually set, not for ordinary edits in general.
ordinary := req
ordinary.RetentionPurge = false
ordinary.OwnerUserID = recipient.ID
ordinary.Peer = domain.Peer{Type: domain.PeerTypeUser, ID: sender.ID}
if _, err := messages.EditMessage(ctx, ordinary); err != domain.ErrMessageAuthorRequired {
t.Fatalf("non-bypass edit by non-author err = %v, want ErrMessageAuthorRequired", err)
}
}
// TestRetentionPurgeBypassProducesServiceMessage_ChannelMessage is the
// channel counterpart of the private-message test above: channel messages
// store their service action in a structurally separate Action column (see
// tgChannelMessage), so applyRetentionPurgeChannelMessage must clear body/
// media and set Action instead of replacing Media like the private path.
func TestRetentionPurgeBypassProducesServiceMessage_ChannelMessage(t *testing.T) {
pool := testPool(t)
ctx := context.Background()
suffix := randomSuffix(t)
users := NewUserStore(pool)
owner, err := users.Create(ctx, domain.User{AccessHash: 83, Phone: "+1997" + suffix + "03", FirstName: "RetentionChOwner"})
if err != nil {
t.Fatalf("create owner: %v", err)
}
t.Cleanup(func() { _, _ = pool.Exec(ctx, "DELETE FROM users WHERE id = $1", owner.ID) })
channels := NewChannelStore(pool)
created, err := channels.CreateChannel(ctx, domain.CreateChannelRequest{CreatorUserID: owner.ID, Title: "Retention " + suffix, Megagroup: true, Date: 1700003000})
if err != nil {
t.Fatalf("create channel: %v", err)
}
channelID := created.Channel.ID
t.Cleanup(func() { _, _ = pool.Exec(ctx, "DELETE FROM channels WHERE id = $1", channelID) })
sent, err := channels.SendChannelMessage(ctx, domain.SendChannelMessageRequest{
UserID: owner.ID, ChannelID: channelID, RandomID: 771122, Message: "old file",
Media: &domain.MessageMedia{Kind: domain.MessageMediaKindDocument, Document: &domain.Document{ID: time.Now().UnixNano(), AccessHash: 1}},
Date: 1700003001,
})
if err != nil {
t.Fatalf("send channel message: %v", err)
}
req := domain.EditChannelMessageRequest{
ChannelID: channelID,
ID: sent.Message.ID,
RetentionPurge: true,
RetentionPurgeAction: &domain.ChannelMessageAction{Type: domain.ChannelActionCustomText, Text: domain.RetentionPurgeNoticeText},
}
res, err := channels.EditChannelMessage(ctx, req)
if err != nil {
t.Fatalf("EditChannelMessage(RetentionPurge): %v", err)
}
if res.Message.Action == nil || res.Message.Action.Type != domain.ChannelActionCustomText {
t.Fatalf("edited channel message action = %+v, want custom_text", res.Message.Action)
}
if res.Message.Action.Text != domain.RetentionPurgeNoticeText {
t.Fatalf("edited channel message notice text = %q, want %q", res.Message.Action.Text, domain.RetentionPurgeNoticeText)
}
if res.Message.Body != "" || !res.Message.Media.IsZero() {
t.Fatalf("edited channel message body/media not cleared: body=%q media=%+v", res.Message.Body, res.Message.Media)
}
// Idempotency guard: a second retention-purge edit on an already-service
// message must not succeed again.
if _, err := channels.EditChannelMessage(ctx, req); err != domain.ErrMessageNotModified {
t.Fatalf("second RetentionPurge channel edit err = %v, want ErrMessageNotModified", err)
}
// Read-back via ListChannelHistory renders as a service message too.
hist, err := channels.ListChannelHistory(ctx, owner.ID, domain.ChannelHistoryFilter{ChannelID: channelID, Limit: 10})
if err != nil {
t.Fatalf("list channel history: %v", err)
}
var found bool
for _, m := range hist.Messages {
if m.ID == sent.Message.ID {
found = true
if m.Action == nil || m.Action.Type != domain.ChannelActionCustomText {
t.Fatalf("history action = %+v, want custom_text", m.Action)
}
}
}
if !found {
t.Fatalf("purged message %d not found in channel history", sent.Message.ID)
}
}
// TestDocumentCategoryPopulatedOnSend guards Part 2's schema/write-path
// change: sending a document message stamps the already-computed
// domain.DocumentMediaCategory value directly onto the documents row, at the
// same point the shared message_box_media index is written -- zero new
// classification logic.
func TestDocumentCategoryPopulatedOnSend(t *testing.T) {
pool := testPool(t)
ctx := context.Background()
suffix := randomSuffix(t)
users := NewUserStore(pool)
sender, err := users.Create(ctx, domain.User{AccessHash: 84, Phone: "+1997" + suffix + "04", FirstName: "CatSender"})
if err != nil {
t.Fatalf("create sender: %v", err)
}
recipient, err := users.Create(ctx, domain.User{AccessHash: 85, Phone: "+1997" + suffix + "05", FirstName: "CatRecipient"})
if err != nil {
t.Fatalf("create recipient: %v", err)
}
ids := []int64{sender.ID, recipient.ID}
docID := time.Now().UnixNano()
t.Cleanup(func() {
_, _ = pool.Exec(ctx, "DELETE FROM message_boxes WHERE owner_user_id = ANY($1::bigint[])", ids)
_, _ = pool.Exec(ctx, "DELETE FROM private_messages WHERE sender_user_id = ANY($1::bigint[])", ids)
_, _ = pool.Exec(ctx, "DELETE FROM dialogs WHERE user_id = ANY($1::bigint[])", ids)
_, _ = pool.Exec(ctx, "DELETE FROM users WHERE id = ANY($1::bigint[])", ids)
_, _ = pool.Exec(ctx, "DELETE FROM documents WHERE id = $1", docID)
_, _ = pool.Exec(ctx, "DELETE FROM media_references WHERE media_kind = 'document' AND media_id = $1", docID)
})
media := NewMediaStore(pool)
if err := media.PutDocument(ctx, domain.Document{ID: docID, MimeType: "video/mp4", Size: 2048}); err != nil {
t.Fatalf("PutDocument: %v", err)
}
messages := NewMessageStore(pool, WithMessageAllocators(&perUserCounterAllocator{}))
_, err = messages.SendPrivateText(ctx, domain.SendPrivateTextRequest{
SenderUserID: sender.ID,
RecipientUserID: recipient.ID,
RandomID: time.Now().UnixNano(),
Media: &domain.MessageMedia{
Kind: domain.MessageMediaKindDocument,
Document: &domain.Document{
ID: docID, AccessHash: 1, MimeType: "video/mp4",
Attributes: []domain.DocumentAttribute{{Kind: domain.DocAttrVideo, W: 640, H: 480, Duration: 5}},
},
},
Date: int(time.Now().Unix()),
})
if err != nil {
t.Fatalf("send video document: %v", err)
}
var category int16
if err := pool.QueryRow(ctx, "SELECT category FROM documents WHERE id = $1", docID).Scan(&category); err != nil {
t.Fatalf("read document category: %v", err)
}
if domain.MediaCategory(category) != domain.MediaCategoryVideo {
t.Fatalf("document category = %d, want %d (Video)", category, domain.MediaCategoryVideo)
}
}
// TestHardRetentionCategoryFilterOnlyReturnsMatchingCategory guards Part 2's
// query-layer change: ListDocumentIDsForHardRetentionOlderThan now takes a
// category argument and must only return documents in that exact bucket,
// even when another old, blob-owning document sits in a different category.
func TestHardRetentionCategoryFilterOnlyReturnsMatchingCategory(t *testing.T) {
pool := testPool(t)
ctx := context.Background()
media := NewMediaStore(pool)
videoID := time.Now().UnixNano()
musicID := videoID + 1
t.Cleanup(func() {
_, _ = pool.Exec(context.Background(), "DELETE FROM documents WHERE id = ANY($1::bigint[])", []int64{videoID, musicID})
_, _ = pool.Exec(context.Background(), "DELETE FROM file_blobs WHERE location_key = ANY($1::text[])",
[]string{"doc:" + strconv.FormatInt(videoID, 10), "doc:" + strconv.FormatInt(musicID, 10)})
})
for _, d := range []struct {
id int64
category domain.MediaCategory
}{{videoID, domain.MediaCategoryVideo}, {musicID, domain.MediaCategoryMusic}} {
if err := media.PutDocument(ctx, domain.Document{ID: d.id, MimeType: "application/octet-stream", Size: 512}); err != nil {
t.Fatalf("PutDocument %d: %v", d.id, err)
}
if _, err := pool.Exec(ctx, "UPDATE documents SET created_at = now() - interval '100 days', category = $2 WHERE id = $1", d.id, int16(d.category)); err != nil {
t.Fatalf("backdate/categorize document %d: %v", d.id, err)
}
blob := postgresTestBlob("doc:"+strconv.FormatInt(d.id, 10), "cat-filter", 512, "application/octet-stream")
if err := media.PutFileBlob(ctx, blob); err != nil {
t.Fatalf("PutFileBlob %d: %v", d.id, err)
}
}
cutoff := time.Now().Add(-24 * time.Hour)
videoIDs, err := media.ListDocumentIDsForHardRetentionOlderThan(ctx, domain.MediaCategoryVideo, cutoff, 1000)
if err != nil {
t.Fatalf("ListDocumentIDsForHardRetentionOlderThan(Video): %v", err)
}
if !containsInt64(videoIDs, videoID) || containsInt64(videoIDs, musicID) {
t.Fatalf("video-category candidates = %v, want to include %d and exclude %d", videoIDs, videoID, musicID)
}
musicIDs, err := media.ListDocumentIDsForHardRetentionOlderThan(ctx, domain.MediaCategoryMusic, cutoff, 1000)
if err != nil {
t.Fatalf("ListDocumentIDsForHardRetentionOlderThan(Music): %v", err)
}
if !containsInt64(musicIDs, musicID) || containsInt64(musicIDs, videoID) {
t.Fatalf("music-category candidates = %v, want to include %d and exclude %d", musicIDs, musicID, videoID)
}
}
// TestEvictionListsOldestAcrossDocumentsAndPhotos guards Part 3's eviction
// query layer: ListOldestMediaForEviction must return candidates from both
// tables (interleaving by created_at is done by the caller in Go, see
// files.Service.EvictOldestMediaOverBudget) so the oldest-overall item can be
// picked regardless of which table it lives in.
func TestEvictionListsOldestAcrossDocumentsAndPhotos(t *testing.T) {
pool := testPool(t)
ctx := context.Background()
media := NewMediaStore(pool)
docID := time.Now().UnixNano()
photoID := docID + 1
t.Cleanup(func() {
_, _ = pool.Exec(context.Background(), "DELETE FROM documents WHERE id = $1", docID)
_, _ = pool.Exec(context.Background(), "DELETE FROM photos WHERE id = $1", photoID)
_, _ = pool.Exec(context.Background(), "DELETE FROM file_blobs WHERE location_key = ANY($1::text[])",
[]string{"doc:" + strconv.FormatInt(docID, 10), "photo:" + strconv.FormatInt(photoID, 10)})
})
if err := media.PutDocument(ctx, domain.Document{ID: docID, MimeType: "application/octet-stream", Size: 256}); err != nil {
t.Fatalf("PutDocument: %v", err)
}
// The document is the older of the two (200 days vs 100 days) -- it must
// sort before the photo in the merged eviction candidate list.
if _, err := pool.Exec(ctx, "UPDATE documents SET created_at = now() - interval '200 days' WHERE id = $1", docID); err != nil {
t.Fatalf("backdate document: %v", err)
}
if err := media.PutFileBlob(ctx, postgresTestBlob("doc:"+strconv.FormatInt(docID, 10), "evict-doc", 256, "application/octet-stream")); err != nil {
t.Fatalf("PutFileBlob doc: %v", err)
}
if err := media.PutPhoto(ctx, domain.Photo{ID: photoID, AccessHash: 1}); err != nil {
t.Fatalf("PutPhoto: %v", err)
}
if _, err := pool.Exec(ctx, "UPDATE photos SET created_at = now() - interval '100 days' WHERE id = $1", photoID); err != nil {
t.Fatalf("backdate photo: %v", err)
}
if err := media.PutFileBlob(ctx, postgresTestBlob("photo:"+strconv.FormatInt(photoID, 10), "evict-photo", 256, "image/jpeg")); err != nil {
t.Fatalf("PutFileBlob photo: %v", err)
}
candidates, err := media.ListOldestMediaForEviction(ctx, 1000)
if err != nil {
t.Fatalf("ListOldestMediaForEviction: %v", err)
}
var doc, photo *domain.EvictionCandidate
for i := range candidates {
c := candidates[i]
switch {
case c.Kind == domain.MediaKindDocument && c.MediaID == docID:
doc = &candidates[i]
case c.Kind == domain.MediaKindPhoto && c.MediaID == photoID:
photo = &candidates[i]
}
}
if doc == nil || photo == nil {
t.Fatalf("eviction candidates missing doc/photo: %+v", candidates)
}
if !doc.CreatedAt.Before(photo.CreatedAt) {
t.Fatalf("document created_at %v not before photo created_at %v (document should be older)", doc.CreatedAt, photo.CreatedAt)
}
}

View file

@ -825,20 +825,103 @@ func (q *Queries) ListAvailableReactions(ctx context.Context) ([]AvailableReacti
return items, nil
}
const listAvatarOrphanedPhotoIDsOlderThan = `-- name: ListAvatarOrphanedPhotoIDsOlderThan :many
SELECT p.id FROM photos p
WHERE p.orphaned_at IS NOT NULL AND p.orphaned_at < $1::timestamptz
AND EXISTS (SELECT 1 FROM profile_photos pp WHERE pp.photo_id = p.id AND pp.active)
ORDER BY p.orphaned_at ASC
LIMIT $2::int
`
type ListAvatarOrphanedPhotoIDsOlderThanParams struct {
Cutoff pgtype.Timestamptz
BatchLimit int32
}
// Same as ListOrphanedPhotoIDsOlderThan but only photos currently active as
// someone's avatar (profile_photos.active) -- lets the Avatar category carry
// its own retention age. In practice a live avatar is never orphaned (an
// active profile_photos row is itself a media_references entry), so this
// bucket is expected to stay empty under "orphan" mode; kept for symmetry
// with the "hard" mode avatar split, which is the one that actually matters.
func (q *Queries) ListAvatarOrphanedPhotoIDsOlderThan(ctx context.Context, arg ListAvatarOrphanedPhotoIDsOlderThanParams) ([]int64, error) {
rows, err := q.db.Query(ctx, listAvatarOrphanedPhotoIDsOlderThan, arg.Cutoff, arg.BatchLimit)
if err != nil {
return nil, err
}
defer rows.Close()
var items []int64
for rows.Next() {
var id int64
if err := rows.Scan(&id); err != nil {
return nil, err
}
items = append(items, id)
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const listAvatarPhotoIDsForHardRetentionOlderThan = `-- name: ListAvatarPhotoIDsForHardRetentionOlderThan :many
SELECT p.id FROM photos p
WHERE p.created_at < $1::timestamptz
AND EXISTS (SELECT 1 FROM profile_photos pp WHERE pp.photo_id = p.id AND pp.active)
AND EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'photo:' || p.id::text
OR fb.location_key LIKE 'photo:' || p.id::text || ':%'
)
ORDER BY p.created_at ASC
LIMIT $2::int
`
type ListAvatarPhotoIDsForHardRetentionOlderThanParams struct {
Cutoff pgtype.Timestamptz
BatchLimit int32
}
// Same as ListPhotoIDsForHardRetentionOlderThan but only photos currently
// active as someone's avatar (profile_photos.active) -- lets the Avatar
// category carry its own retention age, independent of ordinary shared-media
// photos.
func (q *Queries) ListAvatarPhotoIDsForHardRetentionOlderThan(ctx context.Context, arg ListAvatarPhotoIDsForHardRetentionOlderThanParams) ([]int64, error) {
rows, err := q.db.Query(ctx, listAvatarPhotoIDsForHardRetentionOlderThan, arg.Cutoff, arg.BatchLimit)
if err != nil {
return nil, err
}
defer rows.Close()
var items []int64
for rows.Next() {
var id int64
if err := rows.Scan(&id); err != nil {
return nil, err
}
items = append(items, id)
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const listDocumentIDsForHardRetentionOlderThan = `-- name: ListDocumentIDsForHardRetentionOlderThan :many
SELECT d.id FROM documents d
WHERE d.created_at < $1::timestamptz
AND d.category = $2::smallint
AND EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'doc:' || d.id::text
OR fb.location_key LIKE 'doc:' || d.id::text || ':%'
)
ORDER BY d.created_at ASC
LIMIT $2::int
LIMIT $3::int
`
type ListDocumentIDsForHardRetentionOlderThanParams struct {
Cutoff pgtype.Timestamptz
Category int16
BatchLimit int32
}
@ -848,9 +931,11 @@ type ListDocumentIDsForHardRetentionOlderThanParams struct {
// keeps this sweep from re-selecting the same document forever: once its
// blob bytes are purged (DeleteFileBlobsForDocument removes the file_blobs
// rows but deliberately leaves this documents row in place), it naturally
// drops out of this query on the next pass.
// drops out of this query on the next pass. category scopes to one
// documents.category bucket per sweep tick -- see
// ListOrphanedDocumentIDsOlderThan for the 0/None fallback note.
func (q *Queries) ListDocumentIDsForHardRetentionOlderThan(ctx context.Context, arg ListDocumentIDsForHardRetentionOlderThanParams) ([]int64, error) {
rows, err := q.db.Query(ctx, listDocumentIDsForHardRetentionOlderThan, arg.Cutoff, arg.BatchLimit)
rows, err := q.db.Query(ctx, listDocumentIDsForHardRetentionOlderThan, arg.Cutoff, arg.Category, arg.BatchLimit)
if err != nil {
return nil, err
}
@ -917,20 +1002,156 @@ func (q *Queries) ListFileBlobsByLocationPrefix(ctx context.Context, arg ListFil
return items, nil
}
const listMediaReferences = `-- name: ListMediaReferences :many
SELECT media_kind, media_id, ref_kind, ref_key
FROM media_references
WHERE media_kind = $1::text AND media_id = $2::bigint
`
type ListMediaReferencesParams struct {
MediaKind string
MediaID int64
}
type ListMediaReferencesRow struct {
MediaKind string
MediaID int64
RefKind string
RefKey string
}
// Every live reference to a document/photo -- used by the storage retention
// sweep to turn a hard-retention/eviction blob purge into a visible notice on
// every message that still embeds the purged media (see
// files.notifyRetentionPurge). profile_photo/sticker_set/gift refs have no
// message to edit and are filtered by the caller, not this query.
func (q *Queries) ListMediaReferences(ctx context.Context, arg ListMediaReferencesParams) ([]ListMediaReferencesRow, error) {
rows, err := q.db.Query(ctx, listMediaReferences, arg.MediaKind, arg.MediaID)
if err != nil {
return nil, err
}
defer rows.Close()
var items []ListMediaReferencesRow
for rows.Next() {
var i ListMediaReferencesRow
if err := rows.Scan(
&i.MediaKind,
&i.MediaID,
&i.RefKind,
&i.RefKey,
); err != nil {
return nil, err
}
items = append(items, i)
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const listOldestDocumentsForEviction = `-- name: ListOldestDocumentsForEviction :many
SELECT d.id, d.created_at FROM documents d
WHERE EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'doc:' || d.id::text
OR fb.location_key LIKE 'doc:' || d.id::text || ':%'
)
ORDER BY d.created_at ASC
LIMIT $1::int
`
type ListOldestDocumentsForEvictionRow struct {
ID int64
CreatedAt pgtype.Timestamptz
}
// Active eviction (TELESRV_STORAGE_EVICTION_ENABLE): candidates are every
// document that still owns at least one file_blobs row, oldest-uploaded
// first, REGARDLESS of category or age -- unlike the retention sweeps above,
// eviction only cares about reclaiming bytes once the total physical budget
// (TELESRV_STORAGE_MAX_TOTAL_BYTES) is exceeded. created_at is returned so
// the caller can interleave these with ListOldestPhotosForEviction by actual
// age instead of draining one table before touching the other.
func (q *Queries) ListOldestDocumentsForEviction(ctx context.Context, batchLimit int32) ([]ListOldestDocumentsForEvictionRow, error) {
rows, err := q.db.Query(ctx, listOldestDocumentsForEviction, batchLimit)
if err != nil {
return nil, err
}
defer rows.Close()
var items []ListOldestDocumentsForEvictionRow
for rows.Next() {
var i ListOldestDocumentsForEvictionRow
if err := rows.Scan(&i.ID, &i.CreatedAt); err != nil {
return nil, err
}
items = append(items, i)
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const listOldestPhotosForEviction = `-- name: ListOldestPhotosForEviction :many
SELECT p.id, p.created_at FROM photos p
WHERE EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'photo:' || p.id::text
OR fb.location_key LIKE 'photo:' || p.id::text || ':%'
)
ORDER BY p.created_at ASC
LIMIT $1::int
`
type ListOldestPhotosForEvictionRow struct {
ID int64
CreatedAt pgtype.Timestamptz
}
// See ListOldestDocumentsForEviction -- same oldest-first eviction candidate
// selection, for photos (avatars included: eviction is bytes-only and does
// not honor the Avatar category's separate retention age).
func (q *Queries) ListOldestPhotosForEviction(ctx context.Context, batchLimit int32) ([]ListOldestPhotosForEvictionRow, error) {
rows, err := q.db.Query(ctx, listOldestPhotosForEviction, batchLimit)
if err != nil {
return nil, err
}
defer rows.Close()
var items []ListOldestPhotosForEvictionRow
for rows.Next() {
var i ListOldestPhotosForEvictionRow
if err := rows.Scan(&i.ID, &i.CreatedAt); err != nil {
return nil, err
}
items = append(items, i)
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const listOrphanedDocumentIDsOlderThan = `-- name: ListOrphanedDocumentIDsOlderThan :many
SELECT id FROM documents
WHERE orphaned_at IS NOT NULL AND orphaned_at < $1::timestamptz
AND category = $2::smallint
ORDER BY orphaned_at ASC
LIMIT $2::int
LIMIT $3::int
`
type ListOrphanedDocumentIDsOlderThanParams struct {
Cutoff pgtype.Timestamptz
Category int16
BatchLimit int32
}
// category selects one documents.category bucket per sweep tick (per-category
// retention age, see internal/app/files/retention.go) -- 0 (MediaCategoryNone)
// covers unclassified documents (e.g. stickers), which always use the shared
// global age since there is no per-category override for that bucket.
func (q *Queries) ListOrphanedDocumentIDsOlderThan(ctx context.Context, arg ListOrphanedDocumentIDsOlderThanParams) ([]int64, error) {
rows, err := q.db.Query(ctx, listOrphanedDocumentIDsOlderThan, arg.Cutoff, arg.BatchLimit)
rows, err := q.db.Query(ctx, listOrphanedDocumentIDsOlderThan, arg.Cutoff, arg.Category, arg.BatchLimit)
if err != nil {
return nil, err
}
@ -950,9 +1171,10 @@ func (q *Queries) ListOrphanedDocumentIDsOlderThan(ctx context.Context, arg List
}
const listOrphanedPhotoIDsOlderThan = `-- name: ListOrphanedPhotoIDsOlderThan :many
SELECT id FROM photos
WHERE orphaned_at IS NOT NULL AND orphaned_at < $1::timestamptz
ORDER BY orphaned_at ASC
SELECT p.id FROM photos p
WHERE p.orphaned_at IS NOT NULL AND p.orphaned_at < $1::timestamptz
AND NOT EXISTS (SELECT 1 FROM profile_photos pp WHERE pp.photo_id = p.id AND pp.active)
ORDER BY p.orphaned_at ASC
LIMIT $2::int
`
@ -961,6 +1183,8 @@ type ListOrphanedPhotoIDsOlderThanParams struct {
BatchLimit int32
}
// Excludes photos currently active as someone's avatar -- see
// ListAvatarOrphanedPhotoIDsOlderThan for that split-off bucket.
func (q *Queries) ListOrphanedPhotoIDsOlderThan(ctx context.Context, arg ListOrphanedPhotoIDsOlderThanParams) ([]int64, error) {
rows, err := q.db.Query(ctx, listOrphanedPhotoIDsOlderThan, arg.Cutoff, arg.BatchLimit)
if err != nil {
@ -984,6 +1208,7 @@ func (q *Queries) ListOrphanedPhotoIDsOlderThan(ctx context.Context, arg ListOrp
const listPhotoIDsForHardRetentionOlderThan = `-- name: ListPhotoIDsForHardRetentionOlderThan :many
SELECT p.id FROM photos p
WHERE p.created_at < $1::timestamptz
AND NOT EXISTS (SELECT 1 FROM profile_photos pp WHERE pp.photo_id = p.id AND pp.active)
AND EXISTS (
SELECT 1 FROM file_blobs fb
WHERE fb.location_key = 'photo:' || p.id::text
@ -999,7 +1224,8 @@ type ListPhotoIDsForHardRetentionOlderThanParams struct {
}
// See ListDocumentIDsForHardRetentionOlderThan -- same "hard" retention
// candidate selection, for photos.
// candidate selection, for photos. Excludes photos currently active as
// someone's avatar -- see ListAvatarPhotoIDsForHardRetentionOlderThan.
func (q *Queries) ListPhotoIDsForHardRetentionOlderThan(ctx context.Context, arg ListPhotoIDsForHardRetentionOlderThanParams) ([]int64, error) {
rows, err := q.db.Query(ctx, listPhotoIDsForHardRetentionOlderThan, arg.Cutoff, arg.BatchLimit)
if err != nil {

View file

@ -1252,6 +1252,7 @@ type Document struct {
CreatedAt pgtype.Timestamptz
OwnerUserID int64
OrphanedAt pgtype.Timestamptz
Category int16
}
type EncryptedFile struct {