Limits and gotchas: migrate from Google Memorystore to KiviDB

(ERR key must be valid UTF-8), and RIOT stops at the first one. Before the copy, look for binary key names: redis-cli --scan | python3 -c "import sys…

  • Key names must be valid UTF-8. KiviDB refuses a key whose name is not (ERR key must be valid UTF-8), and RIOT stops at the first one. Before the copy, look for binary key names: redis-cli --scan | python3 -c "import sys; [print(l, end='') for l in sys.stdin.buffer if not l.decode('utf-8','ignore').encode()==l]" | head (use --tls, --user and --pass as for your source). Rename them in the source first.
  • Valkey 9.0 and later. Neither the backup file nor DUMP payloads load into KiviDB 1.2.0; use RIOT --struct, which leaves stream consumer groups and hash field TTLs behind.
  • Writes after an export or backup are not in the file.
  • Keyspace notifications miss FLUSHALL and FLUSHDB. Don't run them on the source during a live copy.
  • IAM tokens expire after an hour and the copy tools don't renew them.
  • RedisShake does not check the server's TLS certificate: the traffic is encrypted, but it does not verify who it talks to. Run it inside your VPC.
  • Cross-region replicas do not come across; copy from the primary.
  • Lua. If the application uses EVAL, EVALSHA or FUNCTION, KiviDB needs Lua (the lua or full build, or Lua scripting on in KiviDB Cloud).