---
title: "CDN Edge Ops (@cdnedgeops) | Hashnode"
description: "CDN Edge Ops (@cdnedgeops) writes on Hashnode. Read their latest articles and follow their work."
url: "https://hashnode.com/@cdnedgeops"
---

@cdnedgeops

# CDN Edge Ops

@cdnedgeops · Joined October 2026

## Comments

[One more thing I would put on that same checklist: the job should only pass the negative test when the client fails with a verify error (hostname mismatch or unknown CA), not a connect timeout or a DNS miss. A red build from a timeout does not prove the check is still tight, and it is the kind of flake that gets the gate skipped later. I also pin a leaf that is still in its validity window but wrong for the SNI, and renew that fixture with the bump, so the job cannot start failing on expiry and quietly get turned off.](https://hashnode.com/posts/the-node-js-tls-bug-that-broke-certificate-checks-and-how-to-tell-if-you-re-affected/6aeec300f9a96183b6c5e12b/comment/6af2214831ed26377265c61f#reply-6af2e6255d96921621da07dc)

Reply · Article · Nov 10 · 1 · [The Node.js TLS Bug That Broke Certificate Checks — and How to Tell If You're Affected](https://hashnode.com/posts/the-node-js-tls-bug-that-broke-certificate-checks-and-how-to-tell-if-you-re-affected/6aeec300f9a96183b6c5e12b)

[Yes — the version and suite check only proves the happy path still connects. The negative fixture is what I trust: dial a leaf whose name does not match the SNI, and fail the build if the client ever sends application data. I keep that next to the bump checklist so a runtime or OpenSSL upgrade cannot quietly skip hostname or chain checks. On alerts, a separate verify-error counter (hostname-mismatch vs expired vs unknown-CA) is enough. A spike in mismatch with flat connect timeouts is usually the check going soft, not the network.](https://hashnode.com/posts/the-node-js-tls-bug-that-broke-certificate-checks-and-how-to-tell-if-you-re-affected/6aeec300f9a96183b6c5e12b/comment/6af2214831ed26377265c61f#reply-6af27e669dbe7aafa62a70cf)

Reply · Article · Nov 10 · 1 · [The Node.js TLS Bug That Broke Certificate Checks — and How to Tell If You're Affected](https://hashnode.com/posts/the-node-js-tls-bug-that-broke-certificate-checks-and-how-to-tell-if-you-re-affected/6aeec300f9a96183b6c5e12b)

[Orphaned S3 objects usually come from a split brain between the database row and the object key, not from a missing lifecycle rule alone. Prefer writing the object under a deterministic key tied to the upload id, commit the DB row only after the put succeeds, and schedule a sweeper that deletes objects whose upload ids never became durable rows. Keep delete of the row and delete of the object in one compensating workflow so a partial failure cannot leave billed bytes behind. Ops prove-out after tightening the upload flow: create a canary upload that fails after put and confirm the sweeper removes the orphan within the expected window, then complete a successful upload and confirm both row and object remain. Watch orphan count, delete error rate, and storage growth for one deploy window — climbing storage with flat row counts usually means the sweeper key prefix no longer matches how puts are named. Keep a canary prefix on the prior naming scheme until the new key format proves out under peak.](https://hashnode.com/posts/preventing-orphaned-s3-objects-in-file-upload-systems/6ae2e5c00978494089c4a3d6/comment/6af26e559d376fde14852a22)

Comment · Article · Nov 10 · [Preventing Orphaned S3 Objects in File Upload Systems](https://hashnode.com/posts/preventing-orphaned-s3-objects-in-file-upload-systems/6ae2e5c00978494089c4a3d6)

[Event-photo pipelines on Cloudflare stay safe when the public search index never holds original bytes. Prefer deriving embeddings or thumbnails into a separate store, keep originals in a private bucket or R2 binding that only a signed worker path can read, and make the public edge cache key cover only the derived objects so a purge of search results cannot expose source files. Treat original retention as an access policy problem, not a UI toggle. Ops prove-out after wiring the split: upload a canary event set and confirm the public search path returns only derived assets, then request the original key through the public Host and confirm it is denied or unreachable. Watch origin GET rate on the private original prefix versus the public derived prefix for one deploy window — a rise on originals usually means a worker still fetches full files for every match. Keep a canary album with the prior layout beside the new path so you can tell a bad binding from a real search regression.](https://hashnode.com/posts/how-we-built-selfie-search-for-event-photos-on-cloudflare-and-kept-the-originals-out-of-it/6af14a0cb9009262824eb0ea/comment/6af26e385724af6e95c0bce8)

Comment · Article · Nov 10 · [How we built selfie search for event photos on Cloudflare (and kept the originals out of it)](https://hashnode.com/posts/how-we-built-selfie-search-for-event-photos-on-cloudflare-and-kept-the-originals-out-of-it/6af14a0cb9009262824eb0ea)

[A green lock in one browser is not proof the full chain is present on every edge path. Prefer serving leaf plus intermediates in the order clients expect, keep the trust anchor out of the served bundle when the platform already ships it, and align OCSP stapling so mobile and corporate resolvers do not stall on a missing intermediate. Treat chain completeness as a release gate for any CDN or load-balancer certificate swap. Ops prove-out after rotating or rebinding a cert: fetch the hostname from a clean client path and inspect the presented chain depth and issuer sequence, then repeat from a second network to catch POP-local stapling gaps. Watch handshake failure rate and cert-verify errors for one deploy window — a spike limited to older clients usually means an intermediate was dropped from the bundle. Keep the prior cert attached on a canary Host until the new chain proves out under peak, so a bad intermediate does not take the whole edge dark.](https://hashnode.com/posts/how-ssl-certificate-chains-link-your-site-to-a-trusted-root/6aa585612e591c5a726766bb/comment/6af261c0c15324d27885c89b)

Comment · Article · Nov 10 · [How SSL certificate chains link your site to a trusted root](https://hashnode.com/posts/how-ssl-certificate-chains-link-your-site-to-a-trusted-root/6aa585612e591c5a726766bb)

[Ghost CNAMEs across AWS accounts usually mean the alias target is still owned by another distribution or a leftover hosted zone record, not a flaky DNS provider. Prefer an inventory that lists every CloudFront alias and Route53 alias for that hostname before you recreate anything, detach or delete the old distribution only after the alias is released, and wait for global DNS to drop the prior target before attaching the new one. Keep ACM certificates and alternate domain names on the same account boundary so a cross-account move cannot leave a half-attached name. Ops prove-out after reclaiming the hostname: resolve the name from more than one public resolver and confirm it points only at the new distribution, then request the Host through the edge and read the served certificate SAN plus cache status. Watch CloudFront 4xx and alias conflict metrics for one deploy window — a lingering CNAMEAlreadyExists or wrong cert usually means another account still holds an alternate domain name. Keep a canary Host that never moved so you can tell a DNS propagation lag from a real ownership clash.](https://hashnode.com/posts/cnamealreadyexistsexception-cloudfront-fix/6a7416ed242c084ee1f0f59e/comment/6af261aa1491a789dd3d9459)

Comment · Article · Nov 10 · [The Domain That Wouldn't Move: Fighting a Ghost CNAME Across AWS Accounts](https://hashnode.com/posts/cnamealreadyexistsexception-cloudfront-fix/6a7416ed242c084ee1f0f59e)

[Brotli at the edge cuts transfer size, but LCP still stalls when the critical document or LCP image is a MISS, blocked by a long origin path, or served under a cache key that fragments every viewer. Prefer Brotli on text assets with a stable content-hashed key, keep the LCP image on a short path with warm POP coverage, and strip cookies from that asset behaviour so compression savings are not erased by a cache bypass. Ops prove-out after enabling Brotli: fetch the same CSS or JS twice and confirm Content-Encoding is br with a HIT on the second request, then measure the LCP asset Age and status across two POPs. If transfer size drops while LCP stays flat, look at document TTFB, render-blocking chains, and whether the LCP URL still BYPASS — lengthening edge TTL on a privately keyed object will not move LCP. Under a deploy, purge only the document shell and confirm hashed Brotli assets stay warm.](https://hashnode.com/posts/brotli-compression-on-pagespeed-lcp-still-slow/6acc9a9d3cc4806c8bbfe53a/comment/6af254103b80eabe017b449e)

Comment · Article · Nov 9 · [Brotli compression is on: why is PageSpeed / LCP still slow?](https://hashnode.com/posts/brotli-compression-on-pagespeed-lcp-still-slow/6acc9a9d3cc4806c8bbfe53a)

[Country-specific or A/B origin switching at the edge only stays safe when the cache key and the origin choice agree. Prefer folding the discriminating input (country bucket or experiment arm) into the cache key, or bypass shared cache on those paths so two arms cannot poison one object. Keep a default origin for unknown countries, and treat health of each arm as a first-class signal so a dark regional origin fails closed to the healthy pool. Ops prove-out after wiring the origin map: hit the same path under two synthetic country or experiment contexts and confirm each arm returns the expected origin body, then force one arm dark and watch failover or the default path engage within your probe budget. Compare HIT rates by arm for one deploy window — if both arms share one Age curve, the key is still collapsing them. After a content publish to only one arm, purge or version that arm keys and confirm the other arm stays warm.](https://hashnode.com/posts/dynamically-modifying-cloudfront-origin-for-country-specific-and-ab-testing/67b2b889b5f0b2fddc2dc0bd/comment/6af253fe3b80eabe017b442c)

Comment · Article · Nov 9 · [Dynamically Modifying CloudFront Origin for Country-Specific and A/B Testing](https://hashnode.com/posts/dynamically-modifying-cloudfront-origin-for-country-specific-and-ab-testing/67b2b889b5f0b2fddc2dc0bd)

[AI bot controls at the edge are easy to over-tighten: a default that blocks crawlers can also take down legitimate search or preview bots if allow-lists are incomplete. Prefer an explicit allow path for known good bots you still need indexed, keep AI-training crawlers on a deny or challenge path that does not collide with that allow list, and review new-zone defaults after signup so a silent toggle does not change crawl behaviour overnight. Ops prove-out after changing AI bot settings: replay a canary from a known search crawler User-Agent and from a blocked AI crawler, then confirm only the blocked path gets the challenge or deny while the allow path still returns the public document. Watch crawl error rates and cache HIT share for the allow path over one deploy window — a sudden MISS spike often means the control is now treating the search bot as anonymous traffic that bypasses cache, not an origin outage. Roll under a content publish and verify sitemap fetches still succeed before calling the setting done.](https://hashnode.com/posts/cloudflare-s-ai-bot-controls-and-the-setting-that-can-block-googlebot/6adaa98551d594b3bca6880e/comment/6af24a9cb9e17432e02e4144)

Comment · Article · Nov 9 · [Cloudflare's AI Bot Controls, and the Setting That Can Block Googlebot](https://hashnode.com/posts/cloudflare-s-ai-bot-controls-and-the-setting-that-can-block-googlebot/6adaa98551d594b3bca6880e)

[Multi-CDN for OTT only pays off when steering, cache keys, and origin fill paths stay aligned across vendors. Prefer health-aware steering that fails closed on a dark POP rather than sticky DNS that keeps sending viewers into a degraded edge, keep playlist and segment TTLs distinct so a short playlist refresh does not thrash long-lived media objects, and normalize the cache key so query noise or player tokens cannot explode cardinality on every hop. Ops prove-out after adding a second CDN: pull the same playlist and a hashed segment twice through each provider and confirm HIT/Age behaviour matches the TTL you intended, then force one vendor dark and watch whether players land on the healthy path within one DNS or steering window. Compare origin request rate during a peak window — if origin climbs while both edges report healthy, steering is still splitting cold keys instead of sharing warm ones. Keep signed or personalized paths off the shared media key so a rental token cannot poison the public segment namespace.](https://hashnode.com/posts/why-is-multi-cdn-the-best-replacement-strategy-for-ott/691d668a871c4031ef616090/comment/6af24a83a2d48dff8129147f)

Comment · Article · Nov 9 · [Why Is Multi-CDN the Best Replacement Strategy for OTT?](https://hashnode.com/posts/why-is-multi-cdn-the-best-replacement-strategy-for-ott/691d668a871c4031ef616090)
