Joined January 2026
137 Photos and videos
Ilia Gusev retweeted
Traditional node lifecycle: apt update weekly. kubelet from one repo, containerd from another. Six months in, two nodes run a different containerd because a repo cached weirdly during one rollout. You find out during an incident. podostack.com/p/talos-kubern…
1
1
8
Traditional node lifecycle: apt update weekly. kubelet from one repo, containerd from another. Six months in, two nodes run a different containerd because a repo cached weirdly during one rollout. You find out during an incident. podostack.com/p/talos-kubern…
1
1
8
No "this node has been patched 47 times and the next one has been patched 49 times." There's the version of the installer image you booted from. That's the node's identity. The reinstall is the cheap operation. The hand-edit is the expensive one, because there is no hand-edit.
1
2
Same shape as Issue #18 at the node layer. What used to be an in-band patching ritual becomes a content-addressed artifact. The cluster pulls one image. The cluster applies it. apt update && reboot becomes one gRPC call. podostack.com/p/talos-kubern…
6
Ilia Gusev retweeted
jsonb feels free. Throw the shape-shifting payload into a column, ship, move on. The bill arrives around month 18: write amplification, GIN index size, schema drift you can't query against. podostack.com/p/postgres-jso…
1
1
3
Ilia Gusev retweeted
Fourteen nodes. Subtly different state from manual fixes nobody owns. The cluster in your Terraform is not the cluster you actually have. The actual one is whatever bytes survived every emergency intervention. podostack.com/p/talos-kubern…
1
1
1
jsonb feels free. Throw the shape-shifting payload into a column, ship, move on. The bill arrives around month 18: write amplification, GIN index size, schema drift you can't query against. podostack.com/p/postgres-jso…
1
1
3
Hybrid is the production answer. A few stable columns (id, tenant_id, created_at, kind, status) plus a metadata jsonb for the long tail. Hot path queries against columns. Cold path digs into metadata. You keep planner statistics, foreign keys, and the option to add constraints.
1
It isn't columns vs jsonb. It's which trade-off the access pattern asks for. Wholesale reads, full-document writes? jsonb. Field predicates, partial updates, real constraints? Columns. podostack.com/p/postgres-jso…
Fourteen nodes. Subtly different state from manual fixes nobody owns. The cluster in your Terraform is not the cluster you actually have. The actual one is whatever bytes survived every emergency intervention. podostack.com/p/talos-kubern…
1
1
1
What you find afterwards is which debugging habits Talos forces you to fix. Ad-hoc log lookups become real observability. Host shells become distroless debug containers. Drift detection crons become "config matches state" assertions that machined enforces continuously.
1
2
Same dynamic as Issue #19. Bound SA tokens failed silently because the legacy assumption was already broken; the cluster compensated for years. Drift is the silent-failure version for nodes. Looks fine until it doesn't. podostack.com/p/talos-kubern…
1