Where Hoichoi's infrastructure should live

Full questionCan LKE Enterprise clusters be created in in-bom-2 (Mumbai 2) today by a new customer, and what is the GA date? Confirm in writing that (a) the creation gate is the Kubernetes Enterprise capability and not ACLP Logs Datacenter LKE-E, and (b) once enrolled, GET /v4/regions on our account returns Kubernetes Enterprise for in-bom-2. Chennai is already excluded on our side — in-maa returns neither flag (F). Why it mattersCategory: blocking the architecture. Without LKE-E in in-bom-2 there is no target cluster region; Chennai is already excluded.

Full questionIs Object Storage E3 at in-bom-2 granted on request to a new account with no billing history, and when does it reach GA? State any SLA that does or does not apply under Limited Availability. (We are not asking whether the endpoint exists — we are asking about grant policy and LA terms.) Why it mattersCategory: blocking the architecture. The object store is the media origin; an LA grant with no SLA is not a production origin.

Full questionOn LKE Enterprise, is the pod CIDR unique per cluster and can we specify it? We need prod and UAT non-overlapping on both pod and service CIDRs, none inside 100.64.0.0/12 [inv R12]. Why it mattersCategory: blocking the architecture. CIDR choices are create-time and irreversible; overlapping prod/UAT CIDRs block peering.

Full questionIs there a supported way to pin a Reserved IPv4 to LKE worker-node egress across recycle? Worker nodes are Linodes and the assignment API takes them; what is missing is LKE-side re-attach automation, and CCM v0.9.8 removed the Cilium BGP path. If no: confirm (a) our account is entitled to Reserved IPs, (b) IP Sharing is supported on instances that also hold a VPC interface, (c) whether Placement Groups can host that pair. 8 partners have our 3 NAT EIPs allow-listed [inv R10], [inv C3]. Why it mattersCategory: blocking the architecture. 8 partners allow-list our 3 NAT EIPs; a drifting egress IP breaks partner integrations.

Full questionWhat are the exact kube-reserved CPU millicores and memory on LKE worker nodes (clusters >= 2026-04-21 reserve "a small amount" with no published figure)? Why it mattersCategory: blocking the architecture. Unpublished kube-reserved makes node sizing and the cost model unverifiable.

Full questionDoes LKE apply any admission policy rejecting system-cluster-critical outside kube-system, or inject a default LimitRange/ResourceQuota? We run 8 such pods and have zero LimitRanges across 192 priority-0 pods [inv R25]. Why it mattersCategory: blocking the architecture. An injected LimitRange or priority-class rejection would break 8 running pods on arrival.

Full questionWhat are the actual IOPS and throughput of local NVMe on g8-dedicated-128-32 and g8-dedicated-64-32? Published nowhere. chi-analytics needs >= 4–5 k IOPS and >= 600 MiB/s sustained [inv R8], and Block Storage at 350 MB/s cannot do it. Why it mattersCategory: blocking the data plane. Block Storage at 350 MB/s cannot host chi-analytics; local NVMe is the only candidate and is unpublished.

Full questionIs the 8,000 IOPS / 350 MB/s figure per-volume or per-Linode aggregate? This decides whether multi-volume striping can rescue chi-analytics. Why it mattersCategory: blocking the data plane. Decides whether mdadm RAID-0 over multiple volumes can reach the chi-analytics requirement.

Full questionDoes Object Storage support If-Match / If-None-Match conditional writes and bulk DeleteObjects? Neither appears in any doc. Our encoder uses If-Match; Redpanda tiered-storage GC uses bulk delete [inv §12.3]. Also: is ExpiredObjectDeleteMarker supported in lifecycle policies? Why it mattersCategory: blocking the data plane. The encoder needs If-Match and Redpanda tiered-storage GC needs bulk delete; each miss is a code change.

Full questionWhat is the typical PUT and range-GET latency for small objects on E3? Redpanda Cloud Topics were tuned against a 35–40 ms operation-latency class [inv R15]. Why it mattersCategory: blocking the data plane. Redpanda Cloud Topics were tuned against a 35–40 ms operation-latency class.

Full questionCan a Block Storage volume be expanded online through the CSI driver on a running node, without powering the Linode off? Why it mattersCategory: blocking the data plane. Offline-only resize means a maintenance window per PVC across 44 prod PVCs.

Full questionWhat per-bucket PUT rate can we be raised to on E3, and how quickly? Default 500/s, documented max 2,000/s (F). 28.4 M media objects at 500/s ~ 16 days; at 2,000/s ~ 4 days. Why it mattersCategory: blocking the data plane. At the 500/s default the 28.4 M-object copy takes ~16 days and does not fit the migration window.

Full questionOn Managed PostgreSQL: what pgvector version ships, what is wal_level by default, and what are the server locale and encoding? We need pgvector >= 0.8, logical replication, and en_US.UTF-8 with glibc collation [inv R16]. Why it mattersCategory: blocking the data plane. Decides Managed PostgreSQL vs falling back to in-cluster CNPG.

Full questionWhat is the per-GB Adaptive Media Delivery price for India at ~3.1 PB/month, and is India banded separately from APAC? Is origin egress from Akamai Object Storage waived or discounted under an AMD contract? Your own docs say outbound is metered even inside the same data center. Must be quoted against the fixed input pack (see why_it_matters); a per-GB quote against an unstated request profile is not a comparable quote. Why it mattersWhole video business case. Fixed pack: 3.145 PB/mo; 4.93 B req/mo; 13.5/21.3/36.2 Gbps mean/median-peak/max; 80.5% India (Kolkata 32.8, Delhi 22.0, Mumbai 8.8, Hyderabad 8.0); 91.9% byte-hit; 0.66-0.84 MB objects; m4s 55.7/ts 22.8/mp4 11.3/webm 10.1. Identical to CF-a.

Full questionHow many PoPs in India, specifically Kolkata, Dhaka, Mumbai, Chennai and Hyderabad? 80 % of our traffic is India; Kolkata alone is 32.8 % [inv §1.3 #1]. Why it mattersCategory: blocking the CDN workstream. Kolkata alone is 32.8 % of bytes; PoP presence there decides delivered quality.

Full questionCan an EdgeWorker on the Basic tier (10 ms CPU, 1.5 MB memory per standard event handler) sustain 4.7 billion HMAC token validations per month, and what do Dynamic/Enterprise tiers cost? Tier limits confirmed (F) — Basic 10 ms CPU / 1.5 MB, Dynamic 20 ms / 2.5 MB, Enterprise 70 ms / 4 MB on onClientRequest and the other standard handlers. A "no, Basic cannot" answer does not close this — Q27 is the follow-on and must be answered in the same reply. Why it mattersCategory: blocking the CDN workstream. The token path is mandatory; a tier that cannot run it is a blocker, not a cost line. Answer with Q27.

Full questionDoes Cloud Wrapper work with an Akamai Object Storage origin, at what price, and can it genuinely not be combined with Tiered Distribution? Why it mattersCategory: blocking the CDN workstream. Origin-shield shape drives origin egress cost on the Akamai design.

Full questionIs Akamai Cloud (Linode) MeitY-empanelled in India? Supply the current SOC 2 report (state Type 1 vs Type 2) and the PCI DSS attestation scope for Cloud Computing Services. Your compliance pages 404 to automated retrieval. This becomes pass/fail rather than a data point if §10.6 I-f returns an India-only residency position (§2.3 A12). Why it mattersCategory: commercial and compliance. Becomes pass/fail if Legal returns an India-only data-residency position (I-f).

Full questionConfirm in writing: is network transfer overage $0.005/GB in Chennai and Mumbai, including NodeBalancer egress? Your API returns $0.005 (F); the APAC pricing page prints $0.015 next to NodeBalancers. Our cost model cannot be signed on a contradiction. Why it mattersCategory: commercial and compliance. A 3x contradiction between the API and the pricing page; the cost model cannot be signed on it.

Full questionIs there capacity in in-bom-2 (and in-maa) for 46 simultaneous g8-dedicated-64-32 instances (1,472 vCPU) on a burst basis, and what is the provisioning latency from zero [inv R5]? Why it mattersCategory: commercial and compliance. Encoder burst today peaks at 25 nodes / 484 cores with a 46-node cap; no capacity means no burst lane.