Thank you for your contribution!~ I will process this PR as soon as possible. 😄

Reviewed locally: go build ./... clean, go vet shows only pre-existing noise, merges cleanly onto current develop (b10c9aa), and go test ./common/ (including the three new EncodeMongoURI cases), ./tunnel/..., ./collector/configure/ all pass. The executor/docsyncer test failures are environmental (no live mongod on localhost:27017), unrelated to this change.

LGTM — no blocking issues. Each change fixes a real defect and the logic is sound:

• sanitize.go: the old sourceConn, _ := ignored the error and then dereferenced sourceConn → nil panic on connect failure; now handled. Good catch.
• doc_syncer.go: registering defer wg.Done() before the connect avoids a genuine wg.Wait() deadlock on the early-return path.
• community_client.go: disconnecting on Connect/Ping failure plugs a client leak, and bounding IsGood()/Disconnect with a 10s timeout matters because IsGood() runs on every batch in ensureConnection.
• operation.go: dropping the client only on network/timeout errors is the right churn fix, and the %v→%w changes in the delete+insert fallbacks are correctly coupled so mongo.IsNetworkError can see through the wrapped error. Misclassified dead connections are still caught by the next IsGood() ping.
• Verified the chain end-to-end: EncodeMongoURI output feeds options.Client().ApplyURI(), so directConnection/replicaSet/maxPoolSize/authSource now reach the driver.

Compatibility with the existing special-char password handling — verified, fully compatible.
The conf template documents that passwords containing @ etc. are URL-encoded (per the Aliyun connection docs). That encoding path (url.UserPassword) is untouched here. Parsing order is what keeps it safe: the userinfo is split first via LastIndex("@"), and only then is the query cut from afterUserInfo — so special chars in the password (which live before the last @) are never affected by the new ? split. I ran the existing special-char cases plus combined ones against both this branch and develop:

All existing TestEncodeMongoURI cases still pass unchanged.

One edge worth noting (pre-existing, not introduced by this PR, out of scope): an unencoded @ inside a query value (e.g. …/db?authSource=ad@min) still mis-splits at LastIndex("@"). I confirmed develop produces the exact same broken output, since the mis-split happens before the new ? cut. Per the URI spec such values must be percent-encoded (authSource=ad%40min), which parses correctly. No action needed here.

A few other minor, non-blocking notes:

1. The commit/PR doesn't reference an issue — recent convention here is fix: ... (#NNNN). Worth linking one.
2. This is a behavior change: query params that were previously dropped silently are now honored. A stray replicaSet, or directConnection=true with multiple hosts, will now surface as a driver error instead of being ignored. Suggest a one-line release note.
3. Minor scope observation: this bundles the query-param feature with ~6 connection-leak/churn fixes. Thematically cohesive (connection lifecycle), so fine to keep, but the feature could arguably stand alone.