Errors And Limits

Direct answers errors in Unipile's shape, with Unipile's status codes and error types, so code written for Unipile's errors keeps working.

{
  "title": "Too many requests",
  "detail": "LinkedIn asked this account to slow down.",
  "type": "errors/too_many_requests",
  "status": 429,
  "retry_after": 900
}

Branch on status and type. detail is for people: its words can change, and it never holds an account's data.

Every Error Type#

Parameter Errors#

A required query parameter that's missing, or one of the wrong type, answers Unipile's 400 too, naming the parameter (never the value you sent):

{ "title": "Missing parameters", "detail": "account_id is required.", "type": "errors/missing_parameters", "status": 400 }

{ "title": "Invalid parameters", "detail": "limit: Input should be a valid integer, unable to parse string as an integer.", "type": "errors/invalid_parameters", "status": 400 }

Cloudflare's Answers#

direct.lempile.com is behind Cloudflare. Two of its answers can reach you instead of Direct's:

• 524: the request took more than 100 seconds (a write waiting its turn, a large upload). For a write, it's an unknown outcome.
• 413: the request body is over Cloudflare's limit (100 MB on most plans): a large video.

Pacing#

Every LinkedIn call of an account goes through that account's own pace:

• The limits are per account, never per key: two apps using one account share its budget.
• One request can cost several LinkedIn calls: a profile with sections, a long page, a write's checks and uploads. Each route's page says what it reads.
• The webhook reads use the same budget (about 300 calls a quiet day, 450 at most; see Webhooks).
• A security check (errors/disconnected_account) is never retried.
• The counters live in Direct's memory and start again when Direct restarts.

Retrying#

Writes#

Dry Runs#

Every write takes ?dry_run=true. Direct does everything up to the write itself (it reads what it checks: the profile before an invitation, the "Message" option before a new chat, the post before its deletion) and answers the plan instead of sending it:

{
  "object": "DryRun",
  "route": "posts.react",
  "method": "POST",
  "path": "/voyager/api/voyagerSocialDashReactions?threadUrn=urn%3Ali%3Aactivity%3A7388000000000000000",
  "body": { "reactionType": "LIKE" },
  "status": 201
}

• Status 200. status is what the real write would answer. uploads, multiphoto and checked appear when the write has files or checks.
• Nothing is sent to a LinkedIn write endpoint, and no file is uploaded. The checks' reads do happen and count in the account's calls.
• Dry runs work for every served account, writes on or off, and don't use an Idempotency-Key.
• A refusal found by the checks (already_connected, no_connection_with_recipient…) answers as the real write would.

Idempotency-Key#

Send an Idempotency-Key header (1 to 200 characters, unique per write) on every real write:

• The key is claimed (per account) before anything is sent. The same key on the same account never writes twice.
• A repeat after the write finished answers the first answer again (same status, same body; an empty 204 stays empty), for 7 days.
• A repeat while the first is still running answers 409 errors/conflict.
• A write that surely didn't happen (a refusal, 429, 503, 500 provider_error) frees the key: the same key may try again. A refusal found by the checks before sending never takes the key at all.
• A write whose outcome is unknown answers the same 504 to every repeat: it never writes again under that key.
• The same key on another route answers 400.

With a key, retrying after a timeout, a 524 or a lost connection is safe: you get the first answer, 409 while it runs, or the write if it never started.

Unknown Outcome#

A write is tried again by Direct only when nothing reached LinkedIn (the proxy refused the connection, or LinkedIn refused the session before reading it). When LinkedIn may have received it (a timeout after sending, or an answer that doesn't name what it made), Direct answers:

{
  "title": "Unknown outcome",
  "detail": "LinkedIn may have received this write; it isn't tried again. Check before sending it once more.",
  "type": "errors/request_timeout",
  "status": 504
}

Check before writing again: the chat's latest messages, the account's posts, the post's comments, the sent invitations.

Other Limits#