antanst/gemini-grc
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix snapshot overwrite logic to preserve successful responses

Browse Source 
- Prevent overwriting snapshots that have valid response codes
- Ensure URL is removed from queue when snapshot update is skipped
- Add last_crawled timestamp tracking for better crawl scheduling
- Remove SkipIdenticalContent flag, simplify content deduplication logic
- Update database schema with last_crawled column and indexes

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
antanst
2025-06-18 11:23:56 +03:00
parent e9d7fa85ff
commit ada6cda4ac
8 changed files with 284 additions and 242 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
TODO.md Normal file
View File

@@ -0,0 +1,38 @@
# TODO
## Outstanding Issues
### 1. Ctrl+C Signal Handling Issue
**Problem**: The crawler sometimes doesn't exit properly when Ctrl+C is pressed.
**Root Cause**: The main thread gets stuck in blocking operations before it can check for signals:
- Database operations in the polling loop (`cmd/crawler/crawler.go:239-250`)
- Job queueing when channel is full (`jobs <- url` can block if workers are slow)
- Long-running database transactions
**Location**: `cmd/crawler/crawler.go` - main polling loop starting at line 233
**Solution**: Add signal/context checking to blocking operations:
- Use cancellable context instead of `context.Background()` for database operations
- Make job queueing non-blocking or context-aware
- Add timeout/cancellation to database operations
### 2. fetchSnapshotsFromHistory() Doesn't Work with --skip-identical-content=true
**Problem**: When `--skip-identical-content=true` (default), URLs with unchanged content get continuously re-queued.
**Root Cause**: The function tracks when content last changed, not when URLs were last crawled:
- Identical content → no new snapshot created
- Query finds old snapshot timestamp → re-queues URL
- Creates infinite loop of re-crawling unchanged content
**Location**: `cmd/crawler/crawler.go:388-470` - `fetchSnapshotsFromHistory()` function
**Solution Options**:
1. Add `last_crawled` timestamp to URLs table
2. Create separate `crawl_attempts` table
3. Always create snapshot entries (even for duplicates) but mark them as such
4. Modify logic to work with existing schema constraints
**Current Status**: Function assumes `SkipIdenticalContent=false` per original comment at line 391.

128
cmd/crawler/crawler.go
View File

@@ -307,139 +307,17 @@ func enqueueSeedURLs(ctx context.Context, tx *sqlx.Tx) error {
return nil
}
//func fetchSnapshotsFromHistory(ctx context.Context, tx *sqlx.Tx, num int, age int) (int, error) {
// // Select <num> snapshots from snapshots table for recrawling
// // They should be at least <age> days old, random order,
// // one snapshot per host if possible.
// historyCtx := contextutil.ContextWithComponent(context.Background(), "fetchSnapshotsFromHistory")
// contextlog.LogDebugWithContext(historyCtx, logging.GetSlogger(), "Looking for %d snapshots at least %d days old to recrawl", num, age)
//
// // Calculate the cutoff date
// cutoffDate := time.Now().AddDate(0, 0, -age)
//
// // SQL to select one random URL per host from snapshots older than the cutoff date
// // TODO implement when gopher enabled
// query := `
// WITH hosts AS (
// SELECT DISTINCT host
// FROM snapshots
// WHERE url ~ '^gemini://[^/]+/?$'
// AND gemtext IS NOT NULL
// AND timestamp < $1
// AND error IS NULL
// ),
// ranked_snapshots AS (
// SELECT
// s.url,
// s.host,
// s.timestamp,
// ROW_NUMBER() OVER (PARTITION BY s.host ORDER BY RANDOM()) as rank
// FROM snapshots s
// JOIN hosts h ON s.host = h.host
// WHERE s.timestamp < $1
// )
// SELECT url, host
// FROM ranked_snapshots
// WHERE rank = 1
// ORDER BY RANDOM()
// LIMIT $2
// `
//
// type SnapshotURL struct {
// URL string `db:"url"`
// Host string `db:"host"`
// }
//
// // Execute the query
// var snapshotURLs []SnapshotURL
// err := tx.Select(&snapshotURLs, query, cutoffDate, num)
// if err != nil {
// return 0, err
// }
//
// if len(snapshotURLs) == 0 {
// historyCtx := contextutil.ContextWithComponent(context.Background(), "fetchSnapshotsFromHistory")
// contextlog.LogInfoWithContext(historyCtx, logging.GetSlogger(), "No old snapshots found to recrawl")
// return 0, nil
// }
//
// // For each selected snapshot, add the URL to the urls table
// insertCount := 0
// for _, snapshot := range snapshotURLs {
// err := gemdb.Database.InsertURL(ctx, tx, snapshot.URL)
// if err != nil {
// logging.LogError("Error inserting URL %s from old snapshot to queue: %v", snapshot.URL, err)
// return 0, err
// }
// insertCount++
// }
//
// // Note: The transaction is committed by the caller (runJobScheduler),
// // not here. This function is called as part of a larger transaction.
// if insertCount > 0 {
// historyCtx := contextutil.ContextWithComponent(context.Background(), "fetchSnapshotsFromHistory")
// contextlog.LogInfoWithContext(historyCtx, logging.GetSlogger(), "Added %d old URLs to recrawl queue", insertCount)
// }
//
// return insertCount, nil
//}
func fetchSnapshotsFromHistory(ctx context.Context, tx *sqlx.Tx, num int, age int) (int, error) {
// Select <num> snapshots from snapshots table for recrawling
// Find URLs where the LATEST crawl attempt (via last_crawled) is at least <age> days old
// Now works correctly with SkipIdenticalContent=true - fixes infinite recrawl loop
// Uses last_crawled timestamp to track actual crawl attempts regardless of content changes
historyCtx := contextutil.ContextWithComponent(context.Background(), "fetchSnapshotsFromHistory")
contextlog.LogDebugWithContext(historyCtx, logging.GetSlogger(), "Looking for %d URLs whose latest crawl attempt is at least %d days old to recrawl", num, age)
// Calculate the cutoff date
cutoffDate := time.Now().AddDate(0, 0, -age)
// SQL to select URLs where the latest crawl attempt is older than the cutoff date
// Now uses last_crawled timestamp to track actual crawl attempts
// Works correctly with SkipIdenticalContent setting - avoids infinite recrawl loop
// One URL per host for diversity, focusing on root domain URLs
query := `
WITH latest_attempts AS (
SELECT
url,
host,
COALESCE(MAX(last_crawled), '1970-01-01'::timestamp) as latest_attempt
FROM snapshots
WHERE url ~ '^gemini://[^/]+/?$'
GROUP BY url, host
),
root_urls_with_content AS (
SELECT DISTINCT
la.url,
la.host,
la.latest_attempt
FROM latest_attempts la
JOIN snapshots s ON s.url = la.url
WHERE (s.gemtext IS NOT NULL OR s.data IS NOT NULL)
AND s.response_code BETWEEN 20 AND 29
),
eligible_urls AS (
SELECT
url,
host,
latest_attempt
FROM root_urls_with_content
WHERE latest_attempt < $1
),
ranked_urls AS (
SELECT
url,
host,
latest_attempt,
ROW_NUMBER() OVER (PARTITION BY host ORDER BY RANDOM()) as rank
FROM eligible_urls
)
SELECT url, host
FROM ranked_urls
WHERE rank = 1
ORDER BY RANDOM()
LIMIT $2
`
// Use the query from db_queries.go to find URLs that need re-crawling
type SnapshotURL struct {
URL string `db:"url"`
@@ -448,7 +326,7 @@ func fetchSnapshotsFromHistory(ctx context.Context, tx *sqlx.Tx, num int, age in
// Execute the query
var snapshotURLs []SnapshotURL
err := tx.Select(&snapshotURLs, query, cutoffDate, num)
err := tx.Select(&snapshotURLs, gemdb.SQL_FETCH_SNAPSHOTS_FROM_HISTORY, cutoffDate, num)
if err != nil {
return 0, err
}

67
common/worker.go
View File

@@ -196,15 +196,15 @@ func WorkOnUrl(ctx context.Context, tx *sqlx.Tx, url string) (err error) {
}
// Check if we should skip a potentially
// identical snapshot
skipIdentical, err := shouldSkipIdenticalSnapshot(ctx, tx, s)
// identical snapshot with one from history
isIdentical, err := isContentIdentical(ctx, tx, s)
if err != nil {
return err
}
if skipIdentical {
contextlog.LogInfoWithContext(ctx, logging.GetSlogger(), "Content identical to existing snapshot, recording crawl attempt")
// Record the crawl attempt to track that we processed this URL
err = gemdb.Database.RecordCrawlAttempt(ctx, tx, s)
if isIdentical {
contextlog.LogInfoWithContext(ctx, logging.GetSlogger(), "Content identical to existing snapshot, updating crawl timestamp")
// Update the last_crawled timestamp to track that we processed this URL
err = gemdb.Database.UpdateLastCrawled(ctx, tx, s.URL.String())
if err != nil {
return err
}
@@ -222,37 +222,46 @@ func WorkOnUrl(ctx context.Context, tx *sqlx.Tx, url string) (err error) {
// Save the snapshot and remove the URL from the queue
if s.Error.ValueOrZero() != "" {
contextlog.LogInfoWithContext(ctx, logging.GetSlogger(), "%2d %s", s.ResponseCode.ValueOrZero(), s.Error.ValueOrZero())
// Only save error if we didn't have any valid
// snapshot data from a previous crawl!
shouldUpdateSnapshot, err := shouldUpdateSnapshotData(ctx, tx, s)
if err != nil {
return err
}
if shouldUpdateSnapshot {
contextlog.LogInfoWithContext(ctx, logging.GetSlogger(), "%2d %s", s.ResponseCode.ValueOrZero(), s.Error.ValueOrZero())
return saveSnapshotAndRemoveURL(ctx, tx, s)
} else {
contextlog.LogInfoWithContext(ctx, logging.GetSlogger(), "%2d %s (but old content exists, not updating)", s.ResponseCode.ValueOrZero(), s.Error.ValueOrZero())
return removeURL(ctx, tx, s.URL.String())
}
} else {
contextlog.LogInfoWithContext(ctx, logging.GetSlogger(), "%2d", s.ResponseCode.ValueOrZero())
return saveSnapshotAndRemoveURL(ctx, tx, s)
}
return saveSnapshotAndRemoveURL(ctx, tx, s)
}
func shouldSkipIdenticalSnapshot(ctx context.Context, tx *sqlx.Tx, s *snapshot.Snapshot) (bool, error) {
// Check if content is identical to previous snapshot and we should skip further processing
if config.CONFIG.SkipIdenticalContent {
identical, err := gemdb.Database.IsContentIdentical(ctx, tx, s)
if err != nil {
return false, err
}
if identical {
return true, nil
}
func shouldUpdateSnapshotData(ctx context.Context, tx *sqlx.Tx, s *snapshot.Snapshot) (bool, error) {
prevSnapshot, err := gemdb.Database.GetLatestSnapshot(ctx, tx, s.URL.String())
if err != nil {
return false, err
}
// We write every Gemini capsule, but still
// skip identical pages that aren't capsules.
if s.MimeType.String != "text/gemini" {
identical, err := gemdb.Database.IsContentIdentical(ctx, tx, s)
if err != nil {
return false, err
}
if identical {
return true, nil
}
if prevSnapshot == nil {
return true, nil
}
if prevSnapshot.ResponseCode.Valid {
return false, nil
}
return true, nil
}
func isContentIdentical(ctx context.Context, tx *sqlx.Tx, s *snapshot.Snapshot) (bool, error) {
// Always check if content is identical to previous snapshot
identical, err := gemdb.Database.IsContentIdentical(ctx, tx, s)
if err != nil {
return false, err
}
return false, nil
return identical, nil
}
// storeLinks checks and stores the snapshot links in the database.

27
config/config.go
View File

@@ -9,19 +9,18 @@ import (
// Config holds the application configuration loaded from environment variables.
type Config struct {
PgURL string
LogLevel slog.Level // Logging level (debug, info, warn, error)
MaxResponseSize int // Maximum size of response in bytes
MaxDbConnections int // Maximum number of database connections.
NumOfWorkers int // Number of concurrent workers
ResponseTimeout int // Timeout for responses in seconds
BlacklistPath string // File that has blacklisted strings of "host:port"
WhitelistPath string // File with URLs that should always be crawled regardless of blacklist
DryRun bool // If false, don't write to disk
GopherEnable bool // Enable Gopher crawling
SeedUrlPath string // Add URLs from file to queue
SkipIdenticalContent bool // When true, skip storing snapshots with identical content
SkipIfUpdatedDays int // Skip re-crawling URLs updated within this many days (0 to disable, default 0)
PgURL string
LogLevel slog.Level // Logging level (debug, info, warn, error)
MaxResponseSize int // Maximum size of response in bytes
MaxDbConnections int // Maximum number of database connections.
NumOfWorkers int // Number of concurrent workers
ResponseTimeout int // Timeout for responses in seconds
BlacklistPath string // File that has blacklisted strings of "host:port"
WhitelistPath string // File with URLs that should always be crawled regardless of blacklist
DryRun bool // If false, don't write to disk
GopherEnable bool // Enable Gopher crawling
SeedUrlPath string // Add URLs from file to queue
SkipIfUpdatedDays int // Skip re-crawling URLs updated within this many days (0 to disable)
}
var CONFIG Config //nolint:gochecknoglobals
@@ -39,7 +38,6 @@ func Initialize() *Config {
maxResponseSize := flag.Int("max-response-size", 1024*1024, "Maximum size of response in bytes")
responseTimeout := flag.Int("response-timeout", 10, "Timeout for network responses in seconds")
blacklistPath := flag.String("blacklist-path", "", "File that has blacklist regexes")
skipIdenticalContent := flag.Bool("skip-identical-content", true, "Skip storing snapshots with identical content")
skipIfUpdatedDays := flag.Int("skip-if-updated-days", 60, "Skip re-crawling URLs updated within this many days (0 to disable)")
whitelistPath := flag.String("whitelist-path", "", "File with URLs that should always be crawled regardless of blacklist")
seedUrlPath := flag.String("seed-url-path", "", "File with seed URLs that should be added to the queue immediatelly")
@@ -56,7 +54,6 @@ func Initialize() *Config {
config.WhitelistPath = *whitelistPath
config.SeedUrlPath = *seedUrlPath
config.MaxDbConnections = *maxDbConnections
config.SkipIdenticalContent = *skipIdenticalContent
config.SkipIfUpdatedDays = *skipIfUpdatedDays
level, err := ParseSlogLevel(*loglevel)

47
db/db.go
View File

@@ -41,7 +41,7 @@ type DbService interface {
// Snapshot methods
SaveSnapshot(ctx context.Context, tx *sqlx.Tx, s *snapshot.Snapshot) error
OverwriteSnapshot(ctx context.Context, tx *sqlx.Tx, s *snapshot.Snapshot) error
RecordCrawlAttempt(ctx context.Context, tx *sqlx.Tx, s *snapshot.Snapshot) error
UpdateLastCrawled(ctx context.Context, tx *sqlx.Tx, url string) error
GetLatestSnapshot(ctx context.Context, tx *sqlx.Tx, url string) (*snapshot.Snapshot, error)
GetSnapshotAtTimestamp(ctx context.Context, tx *sqlx.Tx, url string, timestamp time.Time) (*snapshot.Snapshot, error)
GetAllSnapshotsForURL(ctx context.Context, tx *sqlx.Tx, url string) ([]*snapshot.Snapshot, error)
@@ -374,21 +374,10 @@ func (d *DbServiceImpl) SaveSnapshot(ctx context.Context, tx *sqlx.Tx, s *snapsh
return err
}
// Check if we should skip storing identical content
if config.CONFIG.SkipIdenticalContent {
// Use the context-aware version to check for identical content
identical, err := d.IsContentIdentical(ctx, tx, s)
if err != nil {
return err
} else if identical {
contextlog.LogDebugWithContext(dbCtx, logging.GetSlogger(), "Skipping URL with identical content to existing snapshot: %s", s.URL.String())
return nil
}
}
// Always ensure we have a current timestamp
s.Timestamp = null.TimeFrom(time.Now())
// last_crawled will be set automatically by database DEFAULT
// Always ensure we have current timestamps
currentTime := time.Now()
s.Timestamp = null.TimeFrom(currentTime)
s.LastCrawled = null.TimeFrom(currentTime)
// For PostgreSQL, use the global sqlx.NamedQueryContext function
// The SQL_INSERT_SNAPSHOT already has a RETURNING id clause
@@ -423,26 +412,20 @@ func (d *DbServiceImpl) OverwriteSnapshot(ctx context.Context, tx *sqlx.Tx, s *s
return d.SaveSnapshot(ctx, tx, s)
}
// RecordCrawlAttempt records a crawl attempt without saving full content (when content is identical)
func (d *DbServiceImpl) RecordCrawlAttempt(ctx context.Context, tx *sqlx.Tx, s *snapshot.Snapshot) error {
// UpdateLastCrawled updates the last_crawled timestamp for the most recent snapshot of a URL
func (d *DbServiceImpl) UpdateLastCrawled(ctx context.Context, tx *sqlx.Tx, url string) error {
dbCtx := contextutil.ContextWithComponent(ctx, "database")
contextlog.LogDebugWithContext(dbCtx, logging.GetSlogger(), "Recording crawl attempt for URL %s", s.URL.String())
contextlog.LogDebugWithContext(dbCtx, logging.GetSlogger(), "Updating last_crawled timestamp for URL %s", url)
// Check if the context is cancelled before proceeding
if err := ctx.Err(); err != nil {
return err
}
// Record the crawl attempt with minimal data
// timestamp and last_crawled will be set automatically by database DEFAULT
_, err := tx.ExecContext(ctx, SQL_RECORD_CRAWL_ATTEMPT,
s.URL.String(),
s.Host,
s.MimeType.String,
s.ResponseCode.ValueOrZero(),
s.Error.String)
// Update the last_crawled timestamp for the most recent snapshot
_, err := tx.ExecContext(ctx, SQL_UPDATE_LAST_CRAWLED, url)
if err != nil {
return xerrors.NewError(fmt.Errorf("cannot record crawl attempt for URL %s: %w", s.URL.String(), err), 0, "", true)
return xerrors.NewError(fmt.Errorf("cannot update last_crawled for URL %s: %w", url, err), 0, "", true)
}
return nil
@@ -541,14 +524,6 @@ func (d *DbServiceImpl) IsContentIdentical(ctx context.Context, tx *sqlx.Tx, s *
return false, err
}
// Update: Skipped this because empty pages can be valid
// ex. pages with redirect headers
// Only check for identical content if we have gemtext or data
//if (!s.GemText.Valid || s.GemText.String == "") &&
// (!s.Data.Valid || len(s.Data.V) == 0) {
// return false, nil
//}
// Try to get the latest snapshot for this URL
latestSnapshot := &snapshot.Snapshot{}
err := tx.GetContext(ctx, latestSnapshot, SQL_GET_LATEST_SNAPSHOT, s.URL.String())

97
db/db_queries.go
View File

@@ -67,38 +67,10 @@ LIMIT $1
`
// New query - always insert a new snapshot without conflict handling
SQL_INSERT_SNAPSHOT = `
INSERT INTO snapshots (url, host, timestamp, mimetype, data, gemtext, links, lang, response_code, error, header)
VALUES (:url, :host, :timestamp, :mimetype, :data, :gemtext, :links, :lang, :response_code, :error, :header)
INSERT INTO snapshots (url, host, timestamp, mimetype, data, gemtext, links, lang, response_code, error, header, last_crawled)
VALUES (:url, :host, :timestamp, :mimetype, :data, :gemtext, :links, :lang, :response_code, :error, :header, :last_crawled)
RETURNING id
`
// Keep for backward compatibility, but should be phased out
SQL_INSERT_SNAPSHOT_IF_NEW = `
INSERT INTO snapshots (url, host, timestamp, mimetype, data, gemtext, links, lang, response_code, error, header)
VALUES (:url, :host, :timestamp, :mimetype, :data, :gemtext, :links, :lang, :response_code, :error, :header)
ON CONFLICT DO NOTHING
`
// Update to match the SQL_INSERT_SNAPSHOT - we no longer want to upsert, just insert new versions
SQL_UPSERT_SNAPSHOT = `
INSERT INTO snapshots (url, host, timestamp, mimetype, data, gemtext, links, lang, response_code, error, header)
VALUES (:url, :host, :timestamp, :mimetype, :data, :gemtext, :links, :lang, :response_code, :error, :header)
RETURNING id
`
SQL_UPDATE_SNAPSHOT = `UPDATE snapshots
SET url = :url,
host = :host,
timestamp = :timestamp,
mimetype = :mimetype,
data = :data,
gemtext = :gemtext,
links = :links,
lang = :lang,
response_code = :response_code,
error = :error,
header = :header,
last_crawled = CURRENT_TIMESTAMP
WHERE id = :id
RETURNING id
`
SQL_INSERT_URL = `
INSERT INTO urls (url, host, timestamp)
VALUES (:url, :host, :timestamp)
@@ -115,7 +87,6 @@ RETURNING id
SQL_DELETE_URL = `
DELETE FROM urls WHERE url=$1
`
// New queries for retrieving snapshots
SQL_GET_LATEST_SNAPSHOT = `
SELECT * FROM snapshots
WHERE url = $1
@@ -140,9 +111,65 @@ RETURNING id
AND timestamp BETWEEN $2 AND $3
ORDER BY timestamp DESC
`
// New query to record crawl attempt when content is identical (no new snapshot needed)
SQL_RECORD_CRAWL_ATTEMPT = `
INSERT INTO snapshots (url, host, mimetype, response_code, error)
VALUES ($1, $2, $3, $4, $5)
// Update last_crawled timestamp for the most recent snapshot of a URL
SQL_UPDATE_LAST_CRAWLED = `
UPDATE snapshots
SET last_crawled = CURRENT_TIMESTAMP
WHERE id = (
SELECT id FROM snapshots
WHERE url = $1
ORDER BY timestamp DESC
LIMIT 1
)
`
// SQL_FETCH_SNAPSHOTS_FROM_HISTORY Fetches URLs from snapshots for re-crawling based on last_crawled timestamp
// This query finds root domain URLs that haven't been crawled recently and selects
// one URL per host for diversity. Uses CTEs to:
// 1. Find latest crawl attempt per URL (via MAX(last_crawled))
// 2. Filter to URLs with actual content and successful responses (20-29)
// 3. Select URLs where latest crawl is older than cutoff date
// 4. Rank randomly within each host and pick one URL per host
// Parameters: $1 = cutoff_date, $2 = limit
SQL_FETCH_SNAPSHOTS_FROM_HISTORY = `
WITH latest_attempts AS (
SELECT
url,
host,
COALESCE(MAX(last_crawled), '1970-01-01'::timestamp) as latest_attempt
FROM snapshots
WHERE url ~ '^gemini://[^/]+/?$' AND mimetype = 'text/gemini'
GROUP BY url, host
),
root_urls_with_content AS (
SELECT DISTINCT
la.url,
la.host,
la.latest_attempt
FROM latest_attempts la
JOIN snapshots s ON s.url = la.url
WHERE (s.gemtext IS NOT NULL OR s.data IS NOT NULL)
AND s.response_code BETWEEN 20 AND 29
),
eligible_urls AS (
SELECT
url,
host,
latest_attempt
FROM root_urls_with_content
WHERE latest_attempt < $1
),
ranked_urls AS (
SELECT
url,
host,
latest_attempt,
ROW_NUMBER() OVER (PARTITION BY host ORDER BY RANDOM()) as rank
FROM eligible_urls
)
SELECT url, host
FROM ranked_urls
WHERE rank = 1
ORDER BY RANDOM()
LIMIT $2
`
)

115
misc/sql/cleanup_duplicate_snapshots.sql Normal file
View File

@@ -0,0 +1,115 @@
-- Cleanup script for snapshots table after adding last_crawled column
-- This script consolidates multiple snapshots per URL by:
-- 1. Keeping the latest snapshot with content (non-null gemtext OR data)
-- 2. Setting its last_crawled to the most recent timestamp from any snapshot for that URL
-- 3. Deleting all other snapshots for URLs with multiple snapshots
--
-- IMPORTANT: This script will permanently delete data. Make sure to backup your database first!
BEGIN;
-- Update last_crawled for URLs with multiple snapshots
-- Keep the latest snapshot with content and update its last_crawled to the most recent timestamp
WITH url_snapshots AS (
-- Get all snapshots grouped by URL with row numbers
SELECT
id,
url,
timestamp,
last_crawled,
gemtext,
data,
ROW_NUMBER() OVER (PARTITION BY url ORDER BY timestamp DESC) as rn_by_timestamp
FROM snapshots
),
latest_content_snapshots AS (
-- Find the latest snapshot with content for each URL
SELECT
url,
id as keep_id,
timestamp as keep_timestamp
FROM url_snapshots
WHERE (gemtext IS NOT NULL OR data IS NOT NULL)
AND rn_by_timestamp = (
SELECT MIN(rn_by_timestamp)
FROM url_snapshots us2
WHERE us2.url = url_snapshots.url
AND (us2.gemtext IS NOT NULL OR us2.data IS NOT NULL)
)
),
most_recent_timestamps AS (
-- Get the most recent timestamp (last_crawled or timestamp) for each URL
SELECT
url,
GREATEST(
MAX(timestamp),
COALESCE(MAX(last_crawled), '1970-01-01'::timestamp)
) as most_recent_time
FROM snapshots
GROUP BY url
)
-- Update the last_crawled of snapshots we're keeping
UPDATE snapshots
SET last_crawled = mrt.most_recent_time
FROM latest_content_snapshots lcs
JOIN most_recent_timestamps mrt ON lcs.url = mrt.url
WHERE snapshots.id = lcs.keep_id;
-- Delete all other snapshots for URLs that have multiple snapshots
WITH url_snapshots AS (
SELECT
id,
url,
timestamp,
gemtext,
data,
ROW_NUMBER() OVER (PARTITION BY url ORDER BY timestamp DESC) as rn_by_timestamp
FROM snapshots
),
latest_content_snapshots AS (
-- Find the latest snapshot with content for each URL
SELECT
url,
id as keep_id
FROM url_snapshots
WHERE (gemtext IS NOT NULL OR data IS NOT NULL)
AND rn_by_timestamp = (
SELECT MIN(rn_by_timestamp)
FROM url_snapshots us2
WHERE us2.url = url_snapshots.url
AND (us2.gemtext IS NOT NULL OR us2.data IS NOT NULL)
)
),
snapshots_to_delete AS (
-- Find snapshots to delete (all except the ones we're keeping)
SELECT s.id
FROM snapshots s
LEFT JOIN latest_content_snapshots lcs ON s.id = lcs.keep_id
WHERE lcs.keep_id IS NULL
AND s.url IN (
-- Only for URLs that have multiple snapshots
SELECT url
FROM snapshots
GROUP BY url
HAVING COUNT(*) > 1
)
)
DELETE FROM snapshots
WHERE id IN (SELECT id FROM snapshots_to_delete);
-- Show summary of changes
SELECT
'Cleanup completed. Remaining snapshots: ' || COUNT(*) as summary
FROM snapshots;
-- Show URLs that still have multiple snapshots (should be 0 after cleanup)
SELECT
'URLs with multiple snapshots after cleanup: ' || COUNT(*) as validation
FROM (
SELECT url
FROM snapshots
GROUP BY url
HAVING COUNT(*) > 1
) multi_snapshots;
COMMIT;

5
misc/sql/initdb.sql
View File

@@ -26,7 +26,8 @@ CREATE TABLE snapshots (
lang TEXT,
response_code INTEGER,
error TEXT,
header TEXT
header TEXT,
last_crawled TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT CURRENT_TIMESTAMP
);
CREATE UNIQUE INDEX idx_url_timestamp ON snapshots (url, timestamp);
@@ -41,3 +42,5 @@ CREATE INDEX idx_response_code_error ON snapshots (response_code, error);
CREATE INDEX idx_response_code_error_nulls ON snapshots (response_code, error) WHERE response_code IS NULL AND error IS NULL;
CREATE INDEX idx_snapshots_unprocessed ON snapshots (host) WHERE response_code IS NULL AND error IS NULL;
CREATE INDEX idx_url_latest ON snapshots (url, timestamp DESC);
CREATE INDEX idx_last_crawled ON snapshots (last_crawled);
CREATE INDEX idx_url_last_crawled ON snapshots (url, last_crawled DESC);