WRR is showing a very high CPU cost relative to previous solutions, and
it's unclear why this is.
Add two metrics that should help us see the shape of the subchannel sets
that are being passed to high cost systems in order to confirm/deny
theories.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
De-experiment pick first since we have both affinity and randomness E2E
test running successfully.
---------
Co-authored-by: Yash Tibrewal <yashkt@google.com>
Our current implementation of Join, TryJoin leverage some complicated
template stuff to work, which makes them hard to maintain. I've been
thinking about ways to simplify that for some time and had something
like this in mind - using a code generator that's at least a little more
understandable to code generate most of the complexity into a file that
is checkable.
Concurrently - I have a cool optimization in mind - but it requires that
we can move promises after polling, which is a contract change. I'm
going to work through the set of primitives we have in the coming weeks
and change that contract to enable the optimization.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
<!--
If you know who should review your pull request, please assign it to
that
person, otherwise the pull request would get assigned randomly.
If your pull request is for a specific language, please add the
appropriate
lang label.
-->
Fix sticky-TF behavior such that once we enter TRANSIENT_FAILURE, we do
not leave that state if we get a new address list.
Also, fix handling of subchannels in state TRANSIENT_FAILURE.
Previously, if a subchannel was already in state TRANSIENT_FAILURE when
we wanted to start a connection attempt on it (e.g., because the
subchannel already existed from a different channel, or because it
already existed in the previous subchannel list), we would wait for it
to report IDLE before attempting to connect. This PR changes pick_first
to instead immediately skip the subchannel and move on to the next one.
Now, the only time we wait for a subchannel in TRANSIENT_FAILURE is when
we wrap back around to the first subchannel in the list.
Our current implementation of `Seq`, `TrySeq` leverage some complicated
template stuff to work, which makes them hard to maintain. I've been
thinking about ways to simplify that for some time and had something
like this in mind - using a code generator that's at least a little more
understandable to code generate most of the complexity into a file that
is checkable.
Concurrently - I have a cool optimization in mind - but it requires that
we can move promises after polling, which is a contract change. I'm
going to work through the set of primitives we have in the coming weeks
and change that contract to enable the optimization.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
We decided to not populate `policy_name` with the HTTP filter name in
xDS case. So removing it from `GenerateServiceConfig`. This will be
consistent across languages. The gRFC
[PR](https://github.com/grpc/proposal/pull/346) has been updated.
I have a cool optimization in mind for promises - but it requires that
we can move them after polling, which is a contract change. I'm going to
work through the set of primitives we have in the coming weeks and
change that contract to enable the optimization.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
This PR adds in delegating call tracers that work like so -
If this is the first call tracer that is being added onto the call
context, just add it as earlier.
If this is the second call tracer that is being added onto the call
context, create a delegating call tracer that contains a list of call
tracers (including the first call tracer).
Any more call tracers added, just get added to the list of tracers in
the delegating call tracer.
(This is not yet used other than through tests.)
<!--
If you know who should review your pull request, please assign it to
that
person, otherwise the pull request would get assigned randomly.
If your pull request is for a specific language, please add the
appropriate
lang label.
-->
This doesn't matter for gtest (it does its own sorting later), but for
the fuzzers this will probably save a bit of churn in the corpus and
lead to faster discovery of interesting cases.
There's *very* little difference in cost for a pooled arena allocation
and a tcmalloc allocation - but the pooled allocation causes memory
stranding for call lifetime, whereas the tcmalloc allocation allows that
to be shared between calls - leading to a much lower overall cost.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
Rolls forward https://github.com/grpc/grpc/pull/33871
Second and third commits here fix internal build issues
In particular, add a `// IWYU pragma: no_include <ares_build.h>` since
`ares.h` [includes that
anyways](bad62225b7/include/ares.h (L23))
(and seems unlikely for that to change since it would be breaking)
Implement DNS using dns service for iOS.
Current limitation:
1. Using a custom name server is not supported.
2. Only supports `LookupHostname`. `LookupSRV` and `LookupTXT` are not
implemented.
3. Not tested with single stack (ipv4 or ipv6) environment
4. ~Not tested with multiple ip records per stack~ manually tested with
wsj.com
5. Not tested with multiple interface environment
Need the ability to override server-side keepalive permit without calls
default without affecting client-side settings.
<!--
If you know who should review your pull request, please assign it to
that
person, otherwise the pull request would get assigned randomly.
If your pull request is for a specific language, please add the
appropriate
lang label.
-->
Normally, c-ares related fds are destroyed after all DNS resolution is
finished in [this code
path](c82d31677a/src/core/ext/filters/client_channel/resolver/dns/c_ares/grpc_ares_wrapper.cc (L210)).
Also there are some fds that c-ares may fail to open or write to
initially, and c-ares will close them internally before grpc ever knows
about them.
But if:
1) c-ares opens a socket and successfully writes a request on it
2) then a subsequent read fails
Then c-ares will close the fd in [this code
path](bad62225b7/src/lib/ares_process.c (L740)),
but gRPC will have a reference on the fd and will still use it
afterwards.
Fix here is to leverage the c-ares socket-override API to properly track
fd ownership between c-ares and grpc.
Related: internal issue b/292203138
CNR a WindowsEventEngine listener flake in:
* 10k local Windows development machine runs
* 50k Windows RBE runs
* 10k Windows VM runs
It fails ~5 times per day on the master CI jobs.
This PR adds some logging to try to see if an edge is missed, and
switches the thread pool implementation to see if that makes the flake
go away. If the flakes disappear, I'll try removing one or the other to
see if either independently fix the problem (hopefully not logging).
---------
Co-authored-by: drfloob <drfloob@users.noreply.github.com>
Reverts grpc/grpc#33819
Verified that it passed these jobs:
`grpc/core/master/linux/grpc_basictests_c_cpp_dbg`
`grpc/core/master/linux/grpc_basictests_c_cpp_opt`
`grpc/core/master/linux/grpc_portability`
Why: Cleanup for chttp2_transport ahead of promise conversion - lots of
logic has become interleaved throughout chttp2, so some effort to
isolate logic out is warranted ahead of that conversion.
What: Split configuration and policy tracking for each of ping rate
throttling and abuse detection into their own modules. Add tests for
them.
Incidentally: Split channel args into their own header so that we can
split the policy stuff into separate build targets.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
This PR is expected to fix the flakes of
`//test/core/tsi:ssl_transport_security_test` when built under UBSAN.
Why is this needed? There are several tests in
`ssl_transport_security_test.cc` that involve doing many expensive
operations and PR #33638 recently added one more (namely, repeatedly
signing with an ECDSA key). The slow tests are already altered for MSAN
and TSAN, and now we need to do the same for UBSAN.
Those tests are failing on CIs which do not have twisted installed. Skip
them for now and will fix the docker images next.
<!--
If you know who should review your pull request, please assign it to
that
person, otherwise the pull request would get assigned randomly.
If your pull request is for a specific language, please add the
appropriate
lang label.
-->
This PR implements a c-ares based DNS resolver for EventEngine with the
reference from the original
[grpc_ares_wrapper.h](../blob/master/src/core/ext/filters/client_channel/resolver/dns/c_ares/grpc_ares_wrapper.h).
The PosixEventEngine DNSResolver is implemented on top of that. Tests
which use the client channel resolver API
([resolver.h](../blob/master/src/core/lib/resolver/resolver.h#L54)) are
ported, namely the
[resolver_component_test.cc](../blob/master/test/cpp/naming/resolver_component_test.cc)
and the
[cancel_ares_query_test.cc](../blob/master/test/cpp/naming/cancel_ares_query_test.cc).
The WindowsEventEngine DNSResolver will use the same EventEngine's
grpc_ares_wrapper and will be worked on next.
The
[resolve_address_test.cc](https://github.com/grpc/grpc/blob/master/test/core/iomgr/resolve_address_test.cc)
which uses the iomgr
[DNSResolver](../blob/master/src/core/lib/iomgr/resolve_address.h#L44)
API has been ported to EventEngine's dns_test.cc. That leaves only 2
tests which use iomgr's API, notably the
[dns_resolver_cooldown_test.cc](../blob/master/test/core/client_channel/resolvers/dns_resolver_cooldown_test.cc)
and the
[goaway_server_test.cc](../blob/master/test/core/end2end/goaway_server_test.cc)
which probably need to be restructured to use EventEngine DNSResolver
(for one thing they override the original grpc_ares_wrapper's free
functions). I will try to tackle these in the next step.
<!--
If you know who should review your pull request, please assign it to
that
person, otherwise the pull request would get assigned randomly.
If your pull request is for a specific language, please add the
appropriate
lang label.
-->
Promises code can prevent these bad requests from even reaching the
application, which is beneficial but this test needs a minor update to
handle it.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
Based on the discussion at:
595a75cc5d..e3b402a8fa (r1244325752)
<!--
If you know who should review your pull request, please assign it to
that
person, otherwise the pull request would get assigned randomly.
If your pull request is for a specific language, please add the
appropriate
lang label.
-->
Craig Tiller
Mohan Li
Yousuk Seung
Mark D. Roth
Luwei Ge
Mario Jones Vimal
Hannah Shi
Yash Tibrewal
AJ Heller
apolcyn
Alisha Nanda
Yijie Ma
Matthew Stevenson
Vignesh Babu