The first three vertical posts in this series sorted projects by freshness. PagerDuty alts at 3-of-3 not-alive, Linktree at 2-of-3 not-alive, Datadog at 3-of-3 alive. Useful, but freshness is one axis. This post adds a second: license-dead. The code is shipping, the maintainers are paid, the repo looks healthy in our pill — but the license under it stopped being open-source somewhere along the way, and the self-host story changed shape even though the freshness signal didn't blink.

Four subjects, in chronological order.

The four

How the spine predicted this — and what it got wrong

In the Datadog post we proposed three structural predictors of whether an OSS vertical thrives or decays:

1. Shared data standard
2. Buyer = maintainer
3. Vendor pricing pain universal

A vertical that hits 3-of-3 (observability) compounds. A vertical that hits 0-of-3 (link-in-bio) decays.

The license-drama subjects were supposed to be the awkward case: every one of these scores 3-of-3 on the predictors, and yet the license fork breaks the funding loop. Mongo, Sentry, Elasticsearch, and Redis all have well-documented data shapes (BSON, SDK wire format, ES query DSL, RESP). All four have a strong buyer-maintainer alignment (developers buy, developers maintain). All four have universal pricing pain at the SaaS tier. The structural framing predicted: licenses that prevent the cloud vendor / distro / fork ecosystem from contributing back should poison even otherwise-thriving categories.

That prediction held for two of the four. It missed badly on the other two.

MongoDB (Oct 2018) — spine prediction held

Mongo went SSPL because they were tired of AWS reselling DocumentDB-as-Mongo without contributing back. The license stopped that, but it also got Mongo dropped from Debian, Red Hat, and Fedora. AWS pivoted to "DocumentDB" (a separate codebase with Mongo wire compatibility). Community-driven self-host stagnated. FerretDB, the credible OSS-licensed (Apache-2.0) wire-compatible replacement, didn't ship a usable version until 2022 — four years after the license switch. As of May 2026 it works for most CRUD + aggregation pipelines, but coverage on advanced operators is per-feature. The funding loop genuinely broke; the OSS replacement filled in slowly.

If you were already self-hosting Community Server: keep going. SSPL is fine for internal use. If your compliance regime prohibits non-OSI licenses, FerretDB is the move and mongodump / mongorestore carries you over.

Sentry (Sept 2019, again Oct 2023) — spine prediction held

Sentry first switched from Apache to BSL in 2019, then again to the Functional Source License in 2023. Both moves prevented anyone from running Sentry-as-a-managed-service competing with Sentry SaaS. Self-hosting for your own org's use stayed legal, but the license is no longer OSI-approved and the heavy stack (ClickHouse + Kafka + many services, 16GB RAM minimum) makes that a non-trivial commitment.

The OSS replacement is GlitchTip (AGPL-3.0): wire-compatible with Sentry's SDKs (change the DSN host, no SDK code change), Django + Postgres + Redis, runs on a $5 VPS. It's not "Sentry minus the BSL" — performance/tracing surface is thinner — but for the 80% of teams using Sentry as an error tracker rather than full APM, it's a clean path. As of May 2026 GlitchTip's freshness pill is alive. The category is decayed but functional.

If you were already self-hosting Sentry: also fine. The FSL allows self-host for your own org's use indefinitely. The migration path off, if compliance forces it, is glitchtip-django plus a DSN change.

Elasticsearch (Jan 2021) — spine prediction failed

This is where the framing breaks. Elastic's 2021 license change to dual SSPL/ELv2 was meant to do the same thing Mongo's had: push back on AWS. AWS responded by forking Elasticsearch 7.10.2 (the last Apache release) into OpenSearch within three months, with Apache-2.0, AWS funding, and a coalition of co-maintainers. Five years later, OpenSearch is the lead "Elasticsearch alternative" on every directory, including ours. The cloud vendor became the maintainer.

Elastic themselves quietly added AGPL-3.0 as a third license option in August 2024, partially walking the move back. Both projects ship actively, both have healthy GitHub freshness, and both speak the same query DSL.

The structural prediction (license forks the funding loop, so spine breaks) was wrong here. The funding loop got rerouted through AWS instead of severed. The new structural axis we missed is "cloud-vendor counterweight present" — when a hyperscaler has commercial reason to keep the OSS alive, the license fork doesn't decay the category, it just changes who pays.

If you were already self-hosting Elasticsearch ≤7.10.x: OpenSearch upgrade is wire-compatible. Above 7.10, cross-cluster _reindex over HTTPS works. Meilisearch and Typesense are real options for "site search box" use cases that don't need cluster-grade analytics.

Redis (Mar 2024) — spine prediction failed, faster

Redis switched from BSD-3 to dual RSAL/SSPL in March 2024. The Linux Foundation announced Valkey three weeks later, sponsored by AWS, Google Cloud, Oracle, Ericsson, and Snap. By the end of 2024 every major cloud's managed-Redis offering had migrated to Valkey. Redis themselves added an AGPL-3.0 option in mid-2024, same shape as Elastic's walkback.

Same story as Elasticsearch but compressed in time. Three hyperscalers showed up immediately with funding because the cost of not having an OSS-licensed Redis was higher than the cost of paying maintainers directly.

If you were already self-hosting Redis ≤7.2: switch to Valkey, drop-in compatible. Above 7.4 there is no Redis-only path; Valkey has been the migration target since launch.

The structural axis we were missing

The Datadog post claimed three predictors. License-drama exposed a fourth:

1. Cloud-vendor counterweight present. When a hyperscaler has a managed offering that depends on the OSS, a license-fork triggers a fast OSS-licensed re-fork (OpenSearch, Valkey). When the SaaS company is the only commercial counterweight (Sentry, Mongo), the OSS replacement only emerges from independent communities and takes years to mature.

Under the new spine, license-drama splits in two. The vertical is "license-dead" only when none of the cloud vendors care enough to fund the rescue. Where they do, "license-dead" turns into "license-rerouted" and self-hosters keep a healthy OSS path.

The implication for our directory: a license_change_at field by itself isn't enough — it needs to be paired with a "successor fork health" signal. We've started rendering a small "license changed YYYY" pill on the Sentry, MongoDB Atlas, and Elastic Cloud pages so the signal is visible. Confluent/Redis SaaS pages don't exist on our directory yet — when they're added, they'll get the same treatment.

What's next

This is the fourth vertical-deep post. Coming up:

• The first month-over-month diff (June 2026 — what flipped state across all 100 SaaS pages)
• Form builders post-Typeform (ohmyform abandoned — what's actually shipping)
• A license_change_at retrofit across the rest of the catalogue, plus the "successor fork health" axis added to the per-SaaS pill

If you'd rather read the diff post in your inbox than rediscover us via search, the digest form below ships the same day each post lands.