1 Commits (bcdc807dd34d7a4e2e1dcfdffecd869da66e6efe)

Author	SHA1	Message	Date
Travis Patterson	c4261b19fd	Fix parsing of long queries (#9134) ... Loki uses a `scanner.Scanner` to consume the characters in a query. `lex.go` relies heavily on the assumption a scanner can be copied to perform speculative parsing. The problem is that sometimes speculative parsing overruns the scanners internal buffer of `1024` bytes and incurs an additional read to the underlying reader. At that point, all subsequent scanners are out of sync with their underlying readers and the query may fail to be parsed. In the case of #9132, there is a function call at line `1024`. `lex.go` tries to understand this is a function call by speculatively parsing and incurs an underlying read which consumers the rest of the reader. When the lexer tries to scan the next character with the original scanner, the reader is already consumed even though only 1024 bytes have been scanned The scanner has an internal buffer size of `1024 bytes`. This PR copies `scanner.Scanner` from the `text/scanner` package and increases the internal buffer to `maxInputSize` which guarantees that no calls to `Scan` will incur reads beyond the first one. Ideally we wouldn't pass Scanners around in such a way but the cost of that is to rethink much of Loki's lexing logic. Given we know a max input size and we already have that in memory, this solution should suffice. Fixes https://github.com/grafana/loki/issues/9132	3 years ago