improvement(favicons): resolve platform tenants no provider can see, and retry icons chosen after a timeout

A tenant on a platform whose domain has no icon at either provider, even at the
parent (AWS sign-in portals, Coupa, GitLab Dedicated, Databricks apps, Salesforce,
Okta preview, UKG, Jama), and a self-hosted gitlab./jira./testrail. host, now end
their candidate walk at the platform's own site, after their own host and parents
so a tenant's custom icon still wins. A parent's icon reached only because a more
specific lookup timed out is cached for a day instead of thirty, since the host's
own icon may exist.

Author: waleedlatif1

apps/sim/lib/favicon/hosts.test.ts

@@ -40,6 +40,60 @@ describe('faviconHostCandidates', () => {
     ])
   })
 
+  it.each([
+    ['portal.awsapps.com', ['portal.awsapps.com', 'awsapps.com', 'aws.amazon.com']],
+    ['acme.coupahost.com', ['acme.coupahost.com', 'coupahost.com', 'coupa.com']],
+    [
+      'docs.acme.gitlab-dedicated.site',
+      [
+        'docs.acme.gitlab-dedicated.site',
+        'acme.gitlab-dedicated.site',
+        'gitlab-dedicated.site',
+        'gitlab.com',
+      ],
+    ],
+    [
+      'acme.lightning.force.com',
+      ['acme.lightning.force.com', 'lightning.force.com', 'force.com', 'salesforce.com'],
+    ],
+  ])(
+    'ends a platform tenant %s at the platform, after its own host and parents',
+    (host, expected) => {
+      expect(faviconHostCandidates(host)).toEqual(expected)
+    }
+  )
+
+  it.each([
+    ['gitlab.acme-internal.io', 'gitlab.com'],
+    ['jira.acme-internal.io', 'atlassian.com'],
+    ['testrail.qa.acme-internal.io', 'testrail.com'],
+  ])('ends a self-hosted %s at its product', (host, product) => {
+    expect(faviconHostCandidates(host).at(-1)).toBe(product)
+  })
+
+  it.each([
+    'mygitlab.acme-internal.io',
+    'docs.gitlab.acme-internal.io',
+    'notawsapps.com',
+    'acme.herokuapp.com',
+  ])('adds no platform for %s, which only resembles one', (host) => {
+    const candidates = faviconHostCandidates(host)
+    expect(candidates).not.toContain('gitlab.com')
+    expect(candidates).not.toContain('aws.amazon.com')
+    expect(candidates.length).toBeGreaterThan(0)
+  })
+
+  it.each(['gitlab.corp', 'gitlab', 'jira.local'])(
+    'still offers nothing for internal %s',
+    (host) => {
+      expect(faviconHostCandidates(host)).toEqual([])
+    }
+  )
+
+  it('never repeats a host the walk already reaches', () => {
+    expect(faviconHostCandidates('gitlab.gitlab.com')).toEqual(['gitlab.gitlab.com', 'gitlab.com'])
+  })
+
   it('tries at most three parents', () => {
     expect(faviconHostCandidates('a.b.c.d.e.example.com')).toEqual([
       'a.b.c.d.e.example.com',

apps/sim/lib/favicon/hosts.ts

@@ -1,13 +1,16 @@
 import { parse } from 'tldts'
+import { platformSite } from '@/lib/favicon/platforms'
 
 const MAX_PARENT_HOSTS = 3
 const HOST_SHAPE = /^[^\s/?#@\\]+$/
 
 /**
  * Hosts to ask a favicon provider about, most specific first. A private tenant such as
  * `acme.atlassian.net` is invisible to providers, but its platform is not, so each parent
- * domain follows — at most three, stopping before the ICANN public suffix (`gitlab.io` is a valid last step;
- * `io` and `co.uk` are not). Internal hosts — IP literals, single labels, and names outside
+ * domain follows — at most three, stopping before the ICANN public suffix (`gitlab.io` is a valid
+ * last step; `io` and `co.uk` are not). A host on a platform no provider can see even at the
+ * parent (`x.awsapps.com`, a self-hosted `gitlab.<company>`) ends at the platform's own site, after
+ * its own host and parents so a tenant's custom icon still wins. Internal hosts — IP literals, single labels, and names outside
  * the public DNS — yield nothing, so they never leave the server, and neither does a name with
  * characters DNS can't hold, which could otherwise reshape a provider's URL.
  */
@@ -23,7 +26,9 @@ export function faviconHostCandidates(input: string): string[] {
   if (isIp || !isIcann || !publicSuffix || /(^|\.)arpa$/.test(publicSuffix)) return []
   const labels = hostname.split('.')
   const candidateCount = labels.length - publicSuffix.split('.').length
-  return labels
+  const candidates = labels
     .slice(0, Math.min(candidateCount, MAX_PARENT_HOSTS + 1))
     .map((_, index) => labels.slice(index).join('.'))
+  const platform = platformSite(hostname)
+  return platform && !candidates.includes(platform) ? [...candidates, platform] : candidates
 }

apps/sim/lib/favicon/platforms.ts

@@ -0,0 +1,40 @@
+/**
+ * Platforms whose tenant hosts no icon provider can see — not even at the parent domain — mapped
+ * to the platform's own site, so a tenant still shows the product it is. Generic hosting domains
+ * (`herokuapp.com`, `azurewebsites.net`, `cloudfront.net`) are deliberately absent: they serve
+ * other companies' apps, and the host's logo would misrepresent them.
+ */
+const PLATFORM_SITES: ReadonlyMap<string, string> = new Map([
+  ['awsapps.com', 'aws.amazon.com'],
+  ['azuredatabricks.net', 'databricks.com'],
+  ['cloud.databricks.com', 'databricks.com'],
+  ['coupahost.com', 'coupa.com'],
+  ['databricksapps.com', 'databricks.com'],
+  ['gitlab-dedicated.site', 'gitlab.com'],
+  ['jamacloud.com', 'jamasoftware.com'],
+  ['lightning.force.com', 'salesforce.com'],
+  ['my.salesforce.com', 'salesforce.com'],
+  ['oktapreview.com', 'okta.com'],
+  ['ultipro.com', 'ukg.com'],
+])
+
+/** Self-hosted tools conventionally served at `<product>.<company domain>`. */
+const SELF_HOSTED_PRODUCTS: ReadonlyMap<string, string> = new Map([
+  ['confluence', 'atlassian.com'],
+  ['gitlab', 'gitlab.com'],
+  ['grafana', 'grafana.com'],
+  ['jenkins', 'jenkins.io'],
+  ['jira', 'atlassian.com'],
+  ['sentry', 'sentry.io'],
+  ['testrail', 'testrail.com'],
+])
+
+/** The public site of the platform a host belongs to, matched on whole labels, or `null`. */
+export function platformSite(hostname: string): string | null {
+  const labels = hostname.split('.')
+  for (let index = 0; index < labels.length; index++) {
+    const site = PLATFORM_SITES.get(labels.slice(index).join('.'))
+    if (site) return site
+  }
+  return SELF_HOSTED_PRODUCTS.get(labels[0]) ?? null
+}

apps/sim/lib/favicon/resolve.server.ts

@@ -12,6 +12,8 @@ const PROVIDER_TIMEOUT_MS = 1500
 const RESOLVE_DEADLINE_MS = 4000
 const HIT_TTL_SECONDS = 30 * 24 * 60 * 60
 const MISS_TTL_SECONDS = 24 * 60 * 60
+/** A parent's icon chosen because the host's own lookup timed out: retried a day later, not a month. */
+const PROVISIONAL_HIT_TTL_SECONDS = 24 * 60 * 60
 const UNAVAILABLE_TTL_SECONDS = 5 * 60
 /** Cached values: an icon's bytes, or one of these markers. */
 const MISS_MARKER = ''
@@ -28,7 +30,7 @@ const PROVIDERS: readonly ((host: string) => string)[] = [
 ]
 
 type FaviconResolution =
-  | { kind: 'hit'; contentType: string; bytes: Buffer }
+  | { kind: 'hit'; contentType: string; bytes: Buffer; provisional?: true }
   | { kind: 'miss' }
   | { kind: 'unavailable' }
 
@@ -57,7 +59,10 @@ async function writeCache(host: string, resolution: FaviconResolution): Promise<
   if (!redis) return
   const [value, ttl] =
     resolution.kind === 'hit'
-      ? ([resolution.bytes, HIT_TTL_SECONDS] as const)
+      ? ([
+          resolution.bytes,
+          resolution.provisional ? PROVISIONAL_HIT_TTL_SECONDS : HIT_TTL_SECONDS,
+        ] as const)
       : resolution.kind === 'miss'
         ? ([MISS_MARKER, MISS_TTL_SECONDS] as const)
         : ([UNAVAILABLE_MARKER, UNAVAILABLE_TTL_SECONDS] as const)
@@ -101,18 +106,21 @@ async function askProviders(host: string, signal: AbortSignal): Promise<FaviconR
 
 /**
  * The most specific host first, so a tenant's own icon beats its platform's. A parent already
- * resolved for a sibling subdomain is reused from the cache instead of asked again.
+ * resolved for a sibling subdomain is reused from the cache instead of asked again. A parent's
+ * icon reached only because a more specific lookup was unavailable is `provisional`: the host's
+ * own icon may exist, so it is not remembered for long.
  */
 async function walk(candidates: string[], signal: AbortSignal): Promise<FaviconResolution> {
   let unavailable = false
   for (const [index, host] of candidates.entries()) {
     const parent = index > 0
     const cached = parent ? await readCache(host) : null
-    if (cached) return cached.kind === 'miss' && unavailable ? { kind: 'unavailable' } : cached
+    if (cached?.kind === 'hit') return unavailable ? { ...cached, provisional: true } : cached
+    if (cached) return unavailable ? { kind: 'unavailable' } : cached
     const result = await askProviders(host, signal)
     if (result.kind === 'hit') {
       if (parent) await writeCache(host, result)
-      return result
+      return unavailable ? { ...result, provisional: true } : result
     }
     if (result.kind === 'unavailable') unavailable = true
   }