Architecture, Connecting a cluster: a restart is put back across, and the dashboard says why a drop waits (#641, #642)
Failure check: only failures a hidden index can cause count; the ring's reach and how to size it (#625)
Tell a cluster about a new privilege only once a check finds it missing (D184 revised, #599)
A privilege a release adds is optional, announced once, and grantable in one step (D184, #599)
Indexterity keeps the MongoDB profiler on for a drop in flight (D183, #596)
Failed-operations check: say why it was skipped, and what a profiler cannot see (#596)
MongoDB 9.0 supported; its $queryStats samples 1% by default (#560)
The peering is a service beside the api, addressed by TUNNEL_PORT (D113)
Architecture: the runtime is a service in the api's own network namespace
rather than a process the api spawns — what loopback buys (no key on a
network, no reachable proxy, no shared secret), what it gives up (a stack
boundary instead of a process boundary), and the URL-with-a-token revision
that came first and was strictly worse.
Running Indexterity: TUNNEL_PORT is the whole of the configuration and
turns on both halves; unset means the feature is off, which is supported.
Names where the container comes from per deployment, and the one
misconfiguration to watch — on with nothing in the pod.
Connecting a cluster: says what to do when the page reports the feature as
turned off.
A tunnel test is on the trail, and an IPv6-only peering works
A stopped pipeline says why, not just that the figures are old
One runtime: our WireGuard implementation is gone, wireguard-go carries the peering
The peering has two runtimes now: ours, and wireguard-go with gvisor
The trail records the VPN peerings too
Connecting a cluster §5.2: testing a tunnel, and editing one
Connecting a cluster, Architecture: VPN tunnels ship as a product feature
#353 landed as a pasted wg0.conf terminated in-process rather than as a sidecar
recipe, so §5.2 becomes instructions for the screen an owner actually uses and
the self-hosted routing recipe moves under it. Architecture records why a shared
sidecar cannot serve the hosted case (one routing table per pod against tenants
on overlapping RFC1918 ranges), and §10.2 states the guard SUBSTITUTION rather
than leaving it as open design work.
link the chart README on dev, as the other repo links do
Connecting a cluster, Architecture: reaching a database over a VPN
Connecting a cluster, Architecture: state the "user database" rule once
Connecting a cluster: the unreadable-database gap is not SQL Server only
PostgreSQL can apply now: the pg_cron route, and the four facts that shape it
Architecture: how a feature module is layered; PostgreSQL apply is parked, not open
Engine ports: name the engine-neutral modules, and fix two dead env vars
Document the fixed scoped-user name and how to remove it
Document the org-level least-privilege rule and the provided/required/redundant panel (#313)
Home says both engines; the connecting page's title matches its contents
Document Query Store: the statements, and what turning it on costs (#246)
Provisioning is not narrowed to the observe selection (#244)
Say what happens when a provisioned login has no user in a database (#244)
Document choosing which databases a cluster is observed on (#244)
Connecting a cluster + Security: SQL Server scoped login (#203)