nil-pointer dereference in getConn

[derfenix]

Have no idea how to handle or investigate this panic.

runtime.errorString: runtime error: invalid memory address or nil pointer dereference
  File "/builds/tt/snami/snami/v2/interfaces/rest/middleware/sentry.go", line 158, in recoverWithSentry
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 80, in (*baseDB).getConn
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 136, in (*baseDB).withConn
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 245, in (*baseDB).exec
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 220, in (*baseDB).ExecContext
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 119, in (*baseDB).initConn
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 84, in (*baseDB).getConn
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 136, in (*baseDB).withConn
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 315, in (*baseDB).query
  File "/builds/tt/snami/snami/.cache/go/pkg/mod/github.com/go-pg/pg/[email protected]/base.go", line 290, in (*baseDB).QueryContext

[derfenix]

Seems like nil conection returned from the pool, but have no idea how it's possible

[elliotcourant]

It looks like there is a code path that might allow both the err and the connection to be nil.

https://github.com/go-pg/pg/blob/e8d0e95cbd9239859683c3b932c520616000edb1/internal/pool/pool.go#L172-L174

If there are more dial errors than the "pool size" then it will return the last dial error.

https://github.com/go-pg/pg/blob/e8d0e95cbd9239859683c3b932c520616000edb1/internal/pool/pool.go#L215-L220

But if the last dial error is nil for whatever reason, then this function could return nil, nil causing such a panic.

---

Looking at this, this seems to be the only way that you could get a nil connection without an error path. But I don't see an immediate way that that error could be nil under those circumstances.

[derfenix]

Found a conn.Execute in onConnect handler in legacy code. May be this thing can lead to the nil conn in case of full pool? Reproduced only on prod, so trying to reproduce it locally first with no luck yet.