ref(attributes): Use JSON for aws.s3.delete & document multi-collection db.collection.name

[pabloDeputter]

Description

Update aws.s3.delete and db.collection.name descriptions and examples.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
sentry-conventions	Ready	Preview	Oct 14, 2026 8:36am UTC

[github-actions]

Semver Impact of This PR

🟢 Patch (bug fixes)

📋 Changelog Preview

This is how your changes will appear in the changelog.
Entries from this PR are highlighted with a left border (blockquote style).

---

New Features ✨

• (attributes) Add http.request.duration by loewenheim in #697

Internal Changes 🔧

• (attributes) Use JSON for aws.s3.delete & document multi-collection db.collection.name by pabloDeputter in #711

• (repo) Update CODEOWNERS to include @giortzisg by Lms24 in #708

---

_{🤖 This preview updates automatically when you update the PR.}

[Lms24]

I looked at documentation in OTel how they treat multiple collections but it doesn't seem like they specify anything regarding multiple collections. Looking at their database span spec, it seems like they only want this attribute to be set, when the db operation is done on one collection.

I think this is free to interpretation but I would assume that when you have something like SELECT * FROM users JOIN orders ON ..., db.collection.name should only hold users.

This is a string attribute, so I'd prefer we avoid joining multiple values into one string.
Are we doing the comma-joined list already somewhere? If so, why the leading ,? 😅

[pabloDeputter]

Yup, I noticed that as well. So for a JOIN OTel will probably omit both users and orders. However db.collection.name is already set to this joined form by Relay; I sent a few queries through the python-SQLAlchemy integration (which never sets db.collection.name): https://sentry-sdks.sentry.io/explore/traces/trace/a4c29d7664f047bfac9ba507460c7ed1/?mode=samples&node=span-a7290e2aed0619ab&project=4511733153660928&source=traces&statsPeriod=1h&table=trace&timestamp=1791970030.188

• join on 2 tables: SELECT * FROM users JOIN orders ON ... => ,orders,users,
• multiple queries: SELECT * FROM "acme-shop-customers" …; SELECT * FROM "acme-shop-orders" … => ,acme-shop-customers,acme-shop-orders,
• single table: SELECT * FROM "Users" WHERE ... => ,users,

So pretty sure Relay sets db.collection.name (when not provided by the SDK) based off query.text; sorts and lower-cases it as well.

So that's why I added the new example to it, I probably should have been a bit more specific on why I changed this.

[pabloDeputter]

sql_tables_from_query() implements this extraction, sorting and lower-casing: https://github.com/getsentry/relay/blob/4e69c4221c841ec7f0c6e43b252bffeb1b3e7dde/relay-event-normalization/src/normalize/span/tag_extraction.rs#L1396-L1431

[pabloDeputter]

So I looked at this again and OTel ([1] remark at https://opentelemetry.io/docs/specs/semconv/registry/attributes/db/#db-collection-name) states that it should not be extracted from db.query.text when the db system supports queries over multiple collections (which basically covers any SQL db), and recommends no case normalization. sql_tables_from_query() does both of these things when db.collection.name isn't set by the SDK (Users => ,users,, users JOIN orders => ,orders,users,).

The commas are inherited from v1 span.domain which was (I think) so a contains-search can differentiate ,users, from ,users_postfix, (getsentry/relay#5399)

[Lms24]

examples takes precedence, so no need to have both keys

Suggested change

	example: '{"Objects":[{"Key":"someFile.yml","VersionId":"3HL4kqtJlcpXroDTDmJ"}],"Quiet":false}',
	examples: ['{"Objects":[{"Key":"someFile.yml","VersionId":"3HL4kqtJlcpXroDTDmJ"}],"Quiet":false}'],
	examples: ['{"Objects":[{"Key":"someFile.yml","VersionId":"3HL4kqtJlcpXroDTDmJ"}],"Quiet":false}'],

[Lms24]

Do you know which SDKs set this attribute and if they all use the JSON-serialized version?

[Lms24]

actually, this is the more important question: #711 (comment)

[pabloDeputter]

In sentry-python (for the next major, so not yet released), boto3 passes it as a dict, so we serialize it as a JSON (https://github.com/getsentry/sentry-python/blob/major/3.0/sentry_sdk/integrations/boto3/_services/s3.py#L70-L79)

[Lms24]

I looked up the OTel definition and the current format is what OTel specifies. Why should we diverge here?

[pabloDeputter]

I'm pretty sure that the Objects[{key=string, ... format is just copied from the AWS CLI docs which is one way of providing it to the CLI, next to JSON.

In the SDK we get the Deletestructure itself, e.g. boto3 passes it as a dict, so we serialize it to a JSON: https://github.com/getsentry/sentry-python/blob/major/3.0/sentry_sdk/integrations/boto3/_services/s3.py#L70-L79 (for the next major, not relased yet)

Other OTel attributes, e.g. aws.dynamodb.* also use JSON-serialized values; so just trying to make it consistent across the aws.* attributes.