> That sounds a bit unsettling. Any theory as to why the defect has not > been manifesting itself with the existing integration tests?
Lack of implementation diversity? Hypothetically, if the same HPACK bug is on the client and the server, they could cancel out, and the tests will pass despite the implementation being non-interoperable and out of spec. The integration tests I'm referring to are hitting a production endpoint for Amazon Kinesis, whose HTTP/2 implementation is based on Netty. > I do not think HPACK can be disabled. By disabling HPACK, I presume you > mean not using Huffman compression? I'll take whatever I can get. The more sources of complexity I can disable, the more code I can rule out (or rule in). I'll see about the approach of marking all the headers sensitive. On Tue, Dec 10, 2019 at 5:06 AM Oleg Kalnichevski <[email protected]> wrote: > On Mon, 2019-12-09 at 15:39 -0800, Ryan Schmitt wrote: > > I'm running some HTTP/2 integration tests and I'm seeing the > > following in > > the wire log: > > > > 15:33:11,667 - c-00000000 << > > "[0xffffffec][0x10][0xffffffe4])a[0xffffffbc][0x6]?:J[0xffffffc1]T[0x > > ffffffb7];" > > 15:33:11,667 - c-00000000 << stream 0 frame: GOAWAY (0x7); flags: > > (0x0); > > length: 35 > > 15:33:11,667 - c-00000000 << Last stream 15 > > 15:33:11,668 - c-00000000 << Code COMPRESSION_ERROR > > 15:33:11,668 - c-00000000 << HPACK - illegal index value > > > > This error is not deterministic, and I've never seen anything quite > > like it > > before. I'm looking for any ideas on how to troubleshoot it. In > > particular, > > I'd like to know whether there's a simple way to disable header > > compression > > on the client, in order to rule out a bug in the HPACK > > implementation. > > That sounds a bit unsettling. Any theory as to why the defect has not > been manifesting itself with the existing integration tests? > > I do not think HPACK can be disabled. By disabling HPACK, I presume you > mean not using Huffman compression? > > One option to disable Huffman compression would be to mark all headers > as sensitive. There is currently no global or connection based switch > to disable the compression but it can be added. > > Oleg > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
