build(deps): Bump redix from 1.5.3 to 1.6.0 in /elixir-orchestration

[dependabot]

Bumps redix from 1.5.3 to 1.6.0.

Changelog

Sourced from redix's changelog.

v1.6.0

• Add support for Redis Cluster. This is a major feature and this its first release, so use with care—I only tested it (extensively) in non-production-grade scenarios; it was tested on various synthetic environments (like with AWS ElastiCache and locally via Docker) but it has not seen the horrors of production traffic. See the Redix.Cluster documentation to get started.
• Add Redix.PubSub.ping/2.
• Fix Elixir 1.19/1.20 warnings.

Unreleased

New features

• Add support for reading from replicas in Redix.Cluster. Start the cluster with read_from_replicas: true to open and supervise a connection to every replica (each one gets a READONLY after connecting), then pass route: :replica or route: :prefer_replica to Redix.Cluster.command/3 and Redix.Cluster.pipeline/3 to send reads to replicas. The default routing stays :primary, so existing behavior is unchanged.
• Route commands outside Redix.Cluster's built-in command table to the correct node. Previously commands the driver didn't recognize were treated as keyless and sent to a random node; now their keys are resolved against the server (via COMMAND INFO, with a COMMAND GETKEYS fallback for commands with movable keys) and the result is cached per command name, so repeated calls add no extra round-trips. The built-in table also grew to cover more commands directly (bit commands, the blocking B* list/sorted-set pops, hash-field expiration commands, XSETID, and BITOP).
• Add a :readonly option to Redix.start_link/1 that issues READONLY after every (re)connection.
• Add a :health_check_interval option to Redix.start_link/1 that detects half-open connections (the socket is open but the server stopped replying) and reconnects. This fixes slow recovery after a Sentinel failover where the old primary is paused (for example via CLIENT PAUSE): the reconnection re-queries the sentinels and lands on the new primary instead of staying wedged. Defaults to :infinity (disabled).

Changes

• Verify SSL server hostnames the way browsers do (RFC 6125) by default, which accepts the wildcard certificates used by servers such as Amazon ElastiCache. This sets customize_hostname_check automatically; you can still override it (or any other default SSL option) through :socket_opts.

Bug fixes

• Bound the total wall-clock of a steady-state Redix.Cluster topology refresh. The refresh runs on the cluster manager's process and probes seed/known nodes serially, so a few black-holed nodes (which accept the connection but never reply) used to block the manager—and every connection restart or MOVED/ASK redirect that needs it—for up to one connection :timeout per node, i.e. tens of seconds. A refresh now stops probing once it exhausts a single :timeout budget, relying on the next refresh for any skipped nodes. The initial topology discovery stays unbounded so startup reliably reaches a reachable seed.

Commits

• See full diff in compare view