Instead of fixing a target size for writes, try to adapt it a little to
observed bandwidth.
The initial algorithm tries to get large writes within 100-1000ms
maximum delay - this range probably wants to be tuned, but let's see.
The hope here is that on slow connections we can not back buffer so much
and so when we need to send a ping-ack it's possible without great
delay.
Just seeing data flowing in after a ping is enough to establish liveness
of a connection, and so we can limit keepalive timeouts to that. Ping
timeouts are necessary for protocol correctness, but may be stuck behind
other traffic, so give them a little more of a grace period.
---------
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 b/304114403
- adds a new experimental tracer useful for diagnosing ping timeout
failures in unit tests
- adds a pair of experimental tracers for fuzzing event engine
- fix the behavior of FuzzingEventEngine so that a RunAfter(0, ...) runs
in the same tick
- up the rate of sends (reduce the send delay) so we guarantee to be
able to send 200kb/sec in fuzzed e2e unit tests
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
The `TickForDuration()` method was using `grpc_core::Timestamp::Now()`
to get the current time, but that was not in sync with the `now_` value
inside the Fuzzing EE itself, with the result that after two subsequent
250ms increments, timers were not being properly fired. I've added a
test that demonstrates this failure without the fix.
Experiment 1: On RST_STREAM: reduce MAX_CONCURRENT_STREAMS for one round
trip.
Experiment 2: If a settings frame is outstanding with a lower
MAX_CONCURRENT_STREAMS than is configured, and we receive a new incoming
stream that would exceed the new cap, randomly reject it.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
Previously chttp2 would allow infinite requests prior to a settings ack
- as the agreed upon limit for requests in that state is infinite.
Instead, after MAX_CONCURRENT_STREAMS requests have been attempted,
start blanket cancelling requests until the settings ack is received.
This can be done efficiently without allocating request state
structures.
This is the initial change of chaotic-good client transport read path,
which is a following PR of the client transport write path at #33876.
There's a pending work of handling endpoint failures in the transport.
It will be added after we have the inter-activity pipe with close
function.
<!--
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.
-->
- Fixes support for the same address being present more than once in the
address list, which was accidentally broken in #34244.
- Change the call attribute to encode the hash as an integer instead of
a string.
This behavior is dangerous because we will crash when the cache is
created, which is not necessarily on application startup and is likely
when you first try to establish an SSL connection. Instead, we log an
error. If the SSL library attempts to put a session ticket in the cache
it will fail to do so, but everything else will continue as normal. In
particular, we will always seamlessly fall back to a full SSL handshake.
Along the way, we also ensure that you cannot put a null `SSL_SESSION`
into the cache, which would lead to a segfault when it is fetched from
the cache.
Previously it turns out it was not safe to run grpc_init in a filter
test - we'd end up mixing event engine implementations, and causing
undefined behavior at grpc_shutdown.
This change makes it safe and fixes a test internally that's flaking at
70% right now (b/302986486).
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
Fix chttp2 too_many_pings test to use only one of IPv4 or IPv6,
depending on test environment.
Also fix dumb reversed conditional bug in some other tests that was
accidentally introduced in #34426.
Looks like we've got a thread race on shutdown with some of these
tests... adding a barrier at the head of tests that require precise
transport counts in order to stabilize.
Isolate ping callback tracking to its own file.
Also takes the opportunity to simplify keepalive code by applying the
ping timeout to all pings.
Adds an experiment to allow multiple pings outstanding too (this was
originally an accidental behavior change of the work, but one that I
think may be useful going forward).
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
More changes as part of the dualstack design:
- Change resolver and LB policy APIs to support multiple addresses per
endpoint. Specifically, replace `ServerAddress` with
`EndpointAddresses`, which encodes more than one address. Per-address
channel args are retained at the same level, so they are now
per-endpoint. For now, `EndpointAddress` provides a single-address ctor
and a single-address accessor for backward compatibility, so
`ServerAdress` is an alias for `EndpointAddresses`; eventually, this
alias and the single-address methods will be removed.
- Add an `EndpointAddressSet` class, which represents an unordered set
of addresses to be used as a map key. This will be used in a number of
LB policies that need to store per-endpoint state.
- Change the LB policy API's `ChannelControlHelper::CreateSubchannel()`
method to take the address and per-endpoint channel args as separate
parameters, so that we don't need to construct a legacy `ServerAddress`
object as we create a new subchannel for each address in the endpoint.
- Change pick_first to flatten the address list.
- Change ring_hash to use `EndpointAddressSet` as the key for its
endpoint map, and to use the first address of the endpoint as the hash
key.
- Change WRR to use `EndpointAddressSet` as the key for its endpoint
weight map.
Note that support for multiple addresses per endpoint is guarded in RR
by the existing `round_robin_delegate_to_pick_fist` experiment and in
WRR by the existing `wrr_delegate_to_pick_first` experiment.
This PR does *not* include support for multiple addresses per endpoint
for the outlier_detection or xds_override_host LB policies; those will
come in subsequent PRs.
Expand our fuzzing capabilities by allowing fuzzers to choose the bits
that go into random number distribution generators.
---------
Co-authored-by: ctiller <ctiller@users.noreply.github.com>
Lets us sever the dependency between stats & exec ctx (finally).
More work likely needs to go into the *mechanism* used here (I'm not a
fan of the per thread index), but that's also something we can address
later.
If the client calls LookupHostname again within the on_resolve callback,
it re-acquires the `request_mu_` before releasing it which results in
deadlock.
With this PR it extracts the request and releases the lock before
calling on_resolve callback so it won't deadlock any more.
Revert the reversion of the SSL_CTX_new change (#34355 reverted #34180 )
with a fix.
There was an issue with using `strcpy` on a `new[] string` in the
constructor of `ssl_credentials`. An ASAN test caught this in some CI
down the line - `ERROR: AddressSanitizer: alloc-dealloc-mismatch
(operator new [] vs free)`
That `strcpy` call was changed to `grp_strdup` which duplicates a string
in a way that can be freed by `gpr_free` and should resolve the ASAN
failure.
Craig Tiller
Mark D. Roth
Yijie Ma
nanahpang
Eugene Ostroukhov
Matthew Stevenson
Yash Tibrewal
Hannah Shi
Gregory Cooke
AJ Heller