The REDIS_CIRCUIT_BREAKER_* settings are read once in resolve into
Resolution::breaker, like get_env_int and get_env_float read them. A value
that does not parse is warned about and keeps its default. Before, a typo
failed connect with InvalidBreakerSetting and put every native store on
per-process memory while RedisCache kept its default breaker.
rustls only reads unencrypted keys, so ssl_password was dropped with a warning
and the TLS handshake failed. The resolver now reads it into TlsOptions and
redis-client decrypts a PKCS#8 ENCRYPTED PRIVATE KEY before handing it to
redis-rs, like the sync redis-py client's load_cert_chain(password=...). A
missing or wrong password, or a legacy Proc-Type encrypted PEM key, fails at
client build with a TlsKey error naming the file
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A Redis setting the Rust client cannot honor is now warned and dropped with
what that means, so a config that works on Python never fails to build on
Rust. health_check_interval, unix_socket_path, ssl_ca_data and ssl_ca_path
are read and applied like redis-py.
health_check_interval resolves to a three-state HealthCheck, so 0 stays off
on a cluster instead of turning on Python's 25 second cluster default, which
only an unset interval takes. unix_socket_path resolves to Target::Unix. A
url's ?db=, ?username= and ?password= are read like parse_url reads them
instead of being dropped, and ssl_min_version no longer warns at TLS 1.2 or
lower, which rustls already meets.
Every command is now the one RedisCache sends. Pipelined scripts go out as
EVALSHA, and a NOSCRIPT op alone runs again after the pipeline like
RedisBatch's _settle_alone. Only a ScriptCall whose reply was NOSCRIPT can be
sent again, so nothing can resend a command that already ran. Increments are
plain INCRBY and INCRBYFLOAT with TTL and EXPIRE in seconds, as
increment_cache, async_increment and RedisBatch send them, instead of Lua.
SET uses EX, and numbers are written the way redis-py encodes them.
