retry and client uuid
A text field created with unique: true refuses a second create with the same value, case and outer spaces ignored, in a batch too. It is the only idempotency key: the API has no header for it.
Partly measured. whether `unique` spans projects, and whether a deleted row's value frees. One project only. Blocked on the work
API
Q After a create times out on the client, how does it learn what committed, and does a client
UUID field, stock or unique, stop the retry from duplicating the row?
Endpoint POST /entity/_batch ; POST /entity/<type> ; POST /schema/<Type>/fields
Docs claim Silent on idempotency and on what a field's unique property enforces.
Actual
POST /entity/_batch, 20 Version creates, client read timeout 10 ms
client raised ReadTimeout after 0.16 s
+0.5 s rows with that code prefix: 0
+1.8 s rows with that code prefix: 20
POST /entity/versions twice, same body, Idempotency-Key and X-Idempotency-Key the same
-> 201, 201: two rows. x-request-id on the response
stock text field Version.description = <uuid>
lookup straight after the 201, ["description", "is", <uuid>] -> 1 row
the same lookup with <uuid> uppercased -> 1 row
the same create again -> 201, 2 rows hold <uuid>
POST /schema/Version/fields text, unique true -> 201, unique {"value": true, "editable": false}
create with <uuid> -> 201
the same create again -> 400
{"status": 400, "code": 104, "title": "Create failed for [Version]: The value for the
zzprobe 127 client uuid field is required to be unique. <br> ",
"meta": {"crud_error_uuid": "<uuid>"}}
<uuid> uppercased -> 400, the same error
" " + <uuid> -> 400, the same error
POST /entity/_batch, one create, <uuid> -> 400, the same error
one batch, two creates, a new uuid -> 400, the same error; rows holding it: 0
Retry after a timeout
| key | duplicate on retry | lookup before retry |
|---|---|---|
| none | yes | nothing to look up |
Idempotency-Key header |
yes, ignored | none |
client UUID in a stock text field (description) |
yes | is finds it at once, case-insensitive |
client UUID in a text field created unique: true |
no: 400 code 104 | is, as above |
Teaches
- A client read timeout says nothing about the server: the abandoned batch committed all 20 rows, readable under two seconds later (recipe 002 saw the same at 1001 rows). Look up before re-sending.
- Create one text field with
unique: truefor the client UUID and send it on every create. The retry of a committed create then fails with code 104 instead of duplicating; treat that 400 as "already there" and read the row by the UUID.uniquecan only be set at create (probe 116). - Uniqueness ignores case and surrounding spaces, so a UUID is safe but a human code is not a fit.
The check runs inside
_batchas well, and a refusal rolls the whole batch back (recipe 002), including two rows of the same batch that collide with each other. - The error title is the field's display name, not its programmatic name, and ends in a literal
<br>. Code 104 is the general create failure, so match on it plus the display name in the title.