Alban Crequy
|
ac5e379da2
|
vendor: update gobpf and tcptracer-bpf
It will help to get missing kretprobe events during the ebpf tracker
initialization and missing connect & accept events, specially visible on
single-cores:
- https://github.com/weaveworks/tcptracer-bpf/pull/36
- https://github.com/iovisor/gobpf/pull/39
It makes use of a Linux patch that should be in Linux >= 4.12
(unreleased yet): https://github.com/iovisor/bcc/issues/1072#issuecomment-289777787
It falls back gracefully on current/older Linux kernels.
The new gobpf vendoring also includes other small fixes such as error
handling.
|
2017-04-06 11:16:25 +02:00 |
|
Iago López Galeiras
|
7541ad9c9c
|
vendor: bump tcptracer-bpf
We found out we were losing kretprobes sometimes because maxactive was set too
low[1]. This problem was more apparent in our GCE test environment because
the kernel was configured with `CONFIG_PREEMPT` not set and we're running
single-core VMs (see https://github.com/weaveworks/tcptracer-bpf/issues/24 for
more details).
Unfortunately, we can't set the maxactive explicitly from userspace. Alban
submitted a kernel patch to allow this[2].
This bumps tcptracer-bpf to include a workaround[3] for this issue in the
guess-offsets phase.
[1]: https://github.com/weaveworks/tcptracer-bpf/issues/24
[2]: https://lkml.org/lkml/2017/3/28/629
[3]: https://github.com/weaveworks/tcptracer-bpf/pull/33
|
2017-03-29 15:57:25 +02:00 |
|
Michael Schubert
|
b49d2e6f9d
|
vendor: update github.com/weaveworks/tcptracer-bpf
|
2017-03-22 10:07:55 +01:00 |
|
Iago López Galeiras
|
6ede5e5c91
|
vendor: add weaveworks/tcptracer-bpf
|
2017-03-08 22:11:12 +01:00 |
|