This skill builds on electric-shapes and electric-proxy-auth. Read those first.

Electric — Debugging Sync Issues

Setup

Enable debug logging to see retry and state machine behavior:

import { ShapeStream, FetchError } from '@electric-sql/client'

const stream = new ShapeStream({
  url: '/api/todos',
  backoffOptions: {
    initialDelay: 1000,
    maxDelay: 32000,
    multiplier: 2,
    debug: true, // Logs retry attempts
  },
  onError: (error) => {
    if (error instanceof FetchError) {
      console.error(`Sync error: ${error.status} at ${error.url}`, error.json)
    }
    return {} // Always return {} to retry
  },
})

Core Patterns

Error retry behavior

Error	Auto-retry?	Action
5xx server errors	Yes (exponential backoff)	Wait and retry
429 rate limit	Yes (respects Retry-After)	Wait and retry
Network errors	Yes (exponential backoff)	Wait and retry
4xx (non-429)	No	Calls onError — return {} to retry manually
409 shape expired	Yes (automatic reset)	Client resets and refetches
MissingHeadersError	Never	Fix CORS/proxy — not retryable even if onError returns {}

Diagnosing MissingHeadersError

This error means Electric response headers (electric-offset, electric-handle) are being stripped, usually by CORS:

MissingHeadersError: This is often due to a proxy not setting CORS correctly
so that all Electric headers can be read by the client.

Fix: expose Electric headers in proxy CORS configuration:

headers.set(
  'Access-Control-Expose-Headers',
  'electric-offset, electric-handle, electric-schema, electric-cursor'
)

Diagnosing fast-loop detection

Console message: "Detected possible fast loop" with diagnostic info.

Cause: proxy/CDN cache key doesn't include handle and offset query params, so the client gets the same stale response repeatedly.

Fix: ensure your proxy/CDN includes all query parameters in its cache key.

For Vercel, add to vercel.json:

{
  "headers": [
    {
      "source": "/api/(.*)",
      "headers": [
        { "key": "CDN-Cache-Control", "value": "no-store" },
        { "key": "Vercel-CDN-Cache-Control", "value": "no-store" }
      ]
    }
  ]
}

Common Mistakes

HIGH Proxy or CDN not including query params in cache key

Wrong:

# nginx caching without query params in key
proxy_cache_key $scheme$host$uri;

Correct:

# Include query params (handle, offset) in cache key
proxy_cache_key $scheme$host$request_uri;

Fast-loop detection fires after 5 requests in 500ms at the same offset. The client auto-clears caches once, then applies backoff, then throws after 5 consecutive detections.

Source: packages/typescript-client/src/client.ts:929-1002

HIGH SSE responses buffered by proxy

Wrong:

location /v1/shape {
  proxy_pass http://electric:3000;
  # Default: proxy_buffering on — SSE responses delayed
}

Correct:

location /v1/shape {
  proxy_pass http://electric:3000;
  proxy_buffering off;
}

For Caddy:

reverse_proxy localhost:3000 {
  flush_interval -1
}

Nginx and Caddy buffer responses by default, causing long delays for SSE live updates. Disable buffering for Electric endpoints. Do NOT disable caching entirely — Electric uses cache headers for request collapsing.

Source: website/docs/guides/troubleshooting.md:69-109

MEDIUM Running 6+ shapes in local dev without HTTP/2

Wrong:

# Running Electric directly on localhost:3000
# With 7+ shapes, browser HTTP/1.1 queues all requests (6 connection limit)

Correct:

# Run Caddy as HTTP/2 proxy on host (not in Docker — Docker prevents HTTP/2)
caddy run --config - --adapter caddyfile <<EOF
localhost:3001 {
  reverse_proxy localhost:3000
}
EOF

Browser HTTP/1.1 limits to 6 TCP connections per origin. With many shapes, requests queue behind each other. Use Caddy as a local HTTP/2 proxy.

Source: website/docs/guides/troubleshooting.md:28-53

HIGH Leaving replication slot active when Electric is stopped

Wrong:

docker stop electric
# Replication slot retains WAL indefinitely — disk fills up

Correct:

docker stop electric

# Drop slot when stopping for extended periods
psql -c "SELECT pg_drop_replication_slot('electric_slot_default');"

# Or set a safety limit
psql -c "ALTER SYSTEM SET max_slot_wal_keep_size = '10GB';"
psql -c "SELECT pg_reload_conf();"

Replication slots retain WAL indefinitely when Electric is disconnected. Postgres disk fills up. Either drop the slot or set max_slot_wal_keep_size.

Source: website/docs/guides/troubleshooting.md:203-316

See also: electric-deployment/SKILL.md — Many sync issues stem from deployment configuration. See also: electric-shapes/SKILL.md — onError semantics and backoff behavior.

Version

Targets @electric-sql/client v1.5.10.