rethink-public/mongo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: close the motor client if an invalid db name raises in __init__ (mongo-9)

Browse Source 
self._client[database] can raise InvalidName after the client is already
built, leaking it via any exception traceback the caller holds. close it
before re-raising, matching the __aenter__ leak fix. also compresses essay
docstrings/comments across the module with no behavior change, keeping the
swallow-and-default contract note and the matched_count/nInserted footguns.

Signed-off-by: disqualifier <dev@disqualifier.me>
This commit is contained in:
dsql
2026-07-02 23:28:04 -04:00
parent 30f0e44097
commit 0a68b65937
4 changed files with 55 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+3 -3
View File
@@ -8,18 +8,18 @@ helpers for the common paths, with a raw escape hatch for everything else.
`requirements.txt`:
```
mongo @ git+ssh://git@git.rethinkstudios.io/rethink-public/mongo.git@v0.1.7
mongo @ git+ssh://git@git.rethinkstudios.io/rethink-public/mongo.git@v0.1.8
```
Direct:
```bash
pip install "mongo @ git+ssh://git@git.rethinkstudios.io/rethink-public/mongo.git@v0.1.7"
pip install "mongo @ git+ssh://git@git.rethinkstudios.io/rethink-public/mongo.git@v0.1.8"
```
Requires `motor` and `pymongo` (pulled transitively).
Drop the `@v0.1.7` suffix from the line above to install the latest unpinned.
Drop the `@v0.1.8` suffix from the line above to install the latest unpinned.
## Usage
pyproject.toml
+1 -1
View File
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
[project]
name = "mongo"
version = "0.1.7"
version = "0.1.8"
description = "async mongodb wrapper over motor with a raw escape hatch"
requires-python = ">=3.10"
dependencies = [
src/mongo/__init__.py
+4 -5
View File
@@ -7,11 +7,10 @@ def __getattr__(name: str):
"""proxy bare package attribute access to the default instance (PEP 562)
lets `import mongo; await mongo.get_documents(...)` work after init().
`from mongo import func` still won't see this (resolved before init).
note: the bare-proxy raw escape hatch is `mongo.collection(name)`, not
`mongo[name]` — module-level subscripting isn't a thing in python, so the proxy
can only forward named attributes. on a Mongo instance both `db[name]` and
`db.collection(name)` work.
`from mongo import func` still won't see this (resolved before init). the
bare-proxy raw escape hatch is `mongo.collection(name)`, not `mongo[name]` —
module-level subscripting isn't a thing in python, so the proxy only forwards
named attributes (an instance's `db[name]` still works fine).
"""
if not name.startswith("_") and hasattr(MongoDB, name):
return getattr(instance(), name)
src/mongo/mongo.py
+46 -59
View File
@@ -12,8 +12,7 @@ object (preferred), one client per process:
async with MongoDB(conn_string, database) as db:
await db.get_documents("users", {})
`Mongo` remains a back-compat alias of `MongoDB` — existing `Mongo(...)` call sites keep
working unchanged.
`Mongo` is a back-compat alias of `MongoDB`; old `Mongo(...)` call sites still work.
module proxy (back-compat), arm once then call bare:
import mongo # not `from mongo import ...`
@@ -21,12 +20,11 @@ module proxy (back-compat), arm once then call bare:
await mongo.get_documents("users", {"active": True})
errors:
wrapped methods log and swallow, returning a safe default
(False / [] / {} / 0 / None). .collection(name) / db[name] return
the raw motor collection: full driver surface, raises, nothing swallowed.
NOTE: mongo SWALLOWS by design — this is the one deliberate difference from the
newer datastore trio (redis/psql/mysql), which is fail-loud. mongo's contract is
kept as-is so existing consumers' branch-on-result control flow doesn't break.
wrapped methods log and swallow, returning a safe default (False / [] / {} / 0 /
None). .collection(name) / db[name] return the raw motor collection: full driver
surface, raises, nothing swallowed. mongo SWALLOWS by design — the one deliberate
difference from the fail-loud redis/psql/mysql trio, kept so existing consumers'
branch-on-result control flow doesn't break.
naming consistency with the trio (all additive — old names still work):
- class is `MongoDB` (was `Mongo`, kept as alias)
@@ -54,23 +52,26 @@ class MongoDB:
"""async mongodb wrapper; one client per process, attach to bot as bot.db"""
def __init__(self, connection_string: str, database: str):
# motor builds the client eagerly here (no network I/O yet). a syntactically
# malformed URI (e.g. "mongodb://") makes the driver raise InvalidURI at
# construction, before connect() — a deploy-time value, so it surfaces at startup
# either way and there is nothing to clean up.
# motor builds the client eagerly (no I/O yet); a bad URI raises InvalidURI
# here with nothing built. an invalid db name raises below, after the client
# exists — close it before propagating so it doesn't outlive the raise.
self._client = AsyncIOMotorClient(connection_string)
try:
self._db = self._client[database]
except BaseException:
self._client.close()
raise
async def connect(self) -> "MongoDB":
"""validate the connection with a ping and return self
motor connects lazily, so this is optional — call it to fail early on a bad
URI/credentials rather than on the first real op (parallel to the trio's
connect()). raises on a bad connection, unlike the swallowing wrapped methods.
optional — motor connects lazily, so this just fails early on a bad
URI/credentials (parallel to the trio's connect()). raises, unlike the
swallowing wrapped methods.
an instance is single-use: pymongo 4.x `close()` is irreversible, and connect()
only pings (it does not rebuild the client). do not reuse an instance after
close()/`async with` exit — construct a fresh MongoDB.
an instance is single-use: pymongo 4.x `close()` is irreversible and
connect() doesn't rebuild the client. construct a fresh MongoDB instead of
reusing one after close()/`async with` exit.
"""
await self._client.admin.command("ping")
return self
@@ -79,10 +80,9 @@ class MongoDB:
try:
return await self.connect()
except BaseException:
# connect()/ping failing here would otherwise leak the motor client (built
# eagerly in __init__ with a background topology monitor + pool) — __aexit__
# is not called when __aenter__ raises. close it before propagating so a
# retry loop against a flapping server doesn't accumulate live clients.
# __aexit__ is not called when __aenter__ raises, so a failed ping here
# would otherwise leak the motor client built in __init__. close it
# before propagating.
self.close()
raise
@@ -122,9 +122,7 @@ class MongoDB:
"""create a collection, optionally seeding ONE index
`index` is a single index key-spec (e.g. `[("a", 1), ("b", 1)]` builds one
compound index, NOT two). for multiple indexes call `create_index` per index.
renamed from `indexes` in v0.1.6 to reflect that it seeds a single index; pass it
positionally, or update a keyword call site from `indexes=` to `index=`.
compound index, NOT two); call `create_index` per index for multiples.
"""
try:
await self._db.create_collection(collection)
@@ -188,11 +186,10 @@ class MongoDB:
) -> int:
"""insert many documents, returning the inserted count
on a partial failure (e.g. ordered=False with one duplicate key among
several documents) the driver raises BulkWriteError AFTER inserting the
non-failing documents; that count is read from `e.details['nInserted']`
so the return value stays honest — a caller retrying the whole batch on a
wrongly-reported 0 would re-insert documents that already landed.
on a partial failure (e.g. ordered=False with one duplicate key) the driver
raises BulkWriteError AFTER inserting the rest; count comes from
`e.details['nInserted']` so a caller retrying on a wrongly-reported 0 doesn't
re-insert documents that already landed.
"""
try:
result = await self._db[collection].insert_many(documents, ordered=ordered)
@@ -379,9 +376,9 @@ class MongoDB:
async def update_document_field(self, collection: str, target: dict, updates: dict) -> bool:
"""$set one or more fields on a single document
returns True when a document matched, even if the write changed nothing (a
`$set` to the identical value leaves `modified_count=0`); use `matched_count`
so an idempotent no-op on an existing doc isn't misread as a failure.
returns True when a document matched, even if `$set` to an identical value
leaves `modified_count=0` — uses `matched_count` so an idempotent no-op isn't
misread as a failure.
"""
try:
response = await self._db[collection].update_one(target, {"$set": updates})
@@ -395,9 +392,8 @@ class MongoDB:
) -> bool:
"""apply raw update operators ($set/$inc/$unset/...) to a single document
returns True when a document matched, even if the operators changed nothing
(e.g. `$set` to an identical value) — use `matched_count` so an idempotent
write on an existing doc isn't misread as a failure.
returns True when a document matched, even if the operators changed nothing —
uses `matched_count` so an idempotent write isn't misread as a failure.
"""
try:
response = await self._db[collection].update_one(target, update_query)
@@ -409,10 +405,9 @@ class MongoDB:
async def upsert_document_field(self, collection: str, target: dict, updates: dict) -> bool:
"""$set fields on a single document, creating it if absent
returns True when a document matched or was upserted, even if the write
changed nothing (a `$set` to the identical value leaves `modified_count=0`);
use `matched_count` so an idempotent no-op upsert on an existing doc isn't
misread as a failure — mirrors update_document_field/update_document_operator.
returns True when a document matched or was upserted, even if `$set` to an
identical value leaves `modified_count=0` — mirrors update_document_field's
matched_count-over-modified_count contract.
"""
try:
response = await self._db[collection].update_one(
@@ -439,9 +434,8 @@ class MongoDB:
) -> bool:
"""$push a value onto an array field
returns True when a document matched (consistent with the other single-doc
update helpers); $push always mutates, so matched_count and modified_count
agree here in practice.
returns True when a document matched; $push always mutates, so matched_count
and modified_count agree here in practice.
"""
try:
response = await self._db[collection].update_one(target, {"$push": {array: value}})
@@ -456,8 +450,7 @@ class MongoDB:
) -> bool:
"""$push to an array and $set a field in one update
returns True when a document matched (consistent with the other single-doc
update helpers).
returns True when a document matched (matched_count, not modified_count).
"""
try:
response = await self._db[collection].update_one(
@@ -474,9 +467,8 @@ class MongoDB:
) -> bool:
"""$pull a value from an array field
returns True when a document matched, even if nothing was pulled (the value
was absent) — use `matched_count` so a no-op pull on an existing doc isn't
misread as a failure.
returns True when a document matched, even if the value was absent and
nothing was pulled — uses `matched_count`, not `modified_count`.
"""
try:
response = await self._db[collection].update_one(target, {"$pull": {array: value}})
@@ -556,10 +548,9 @@ class MongoDB:
# bulk / checks
async def bulk_write(self, collection: str, operations: list, ordered: bool = True) -> dict:
"""
run a batch of pymongo write ops (InsertOne/UpdateOne/DeleteOne/...)
"""run a batch of pymongo write ops (InsertOne/UpdateOne/DeleteOne/...)
caller builds the ops from pymongo, e.g.:
caller builds the ops, e.g.:
from pymongo import UpdateOne
ops = [UpdateOne({'_id': i}, {'$set': {...}}, upsert=True) for i in ids]
returns a summary dict of counts, or an empty dict on failure
@@ -580,8 +571,7 @@ class MongoDB:
async def document_check_multi(
self, collection: str, target: dict, checks: List[str], value: str
) -> bool:
"""
return whether a doc matching target has value in any of the given arrays
"""return whether a doc matching target has value in any of the given arrays
target is a normal filter dict (merged into the query), not a bare _id
"""
@@ -602,12 +592,9 @@ Mongo = MongoDB
# -----------------------------------------------------------------------------
# backwards-compat module proxy
# - lets legacy call sites keep using `await mongo.get_documents(...)`
# without an object, after a one-time `mongo.init(conn, db)`
# - works with `import mongo; mongo.func(...)` (resolved at call time)
# - does NOT work with `from mongo import func` (resolved at import,
# before init runs) — switch those sites to `import mongo`
# backwards-compat module proxy: `import mongo; mongo.init(conn, db)` then
# bare `await mongo.func(...)`. needs `import mongo` — `from mongo import func`
# resolves at import time, before init() runs.
_default: Optional[MongoDB] = None