The ManagedChannelImpl change prevents any LB initialization failure
from producing a useless exception like:
java.lang.NullPointerException
at io.grpc.internal.ManagedChannelImpl.shutdownNameResolverAndLoadBalancer(ManagedChannelImpl.java:321)
at io.grpc.internal.ManagedChannelImpl.panic(ManagedChannelImpl.java:738)
at io.grpc.internal.ManagedChannelImpl$1.uncaughtException(ManagedChannelImpl.java:144)
Instead, now it will have the expected panic behavior of an INTERNAL
Status with a proper cause.
Since the NPE in AutoConfiguredLoadBalancerFactory wouldn't mean much to
users, it now has a more explicit message.
Following the [spec](https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md) on duplicate header names:
**Custom-Metadata** header order is not guaranteed to be preserved except for values with duplicate header names. Duplicate header names may have their values joined with "," as the delimiter and be considered semantically equivalent. Implementations must split Binary-Headers on "," before decoding the Base64-encoded values.
Because otherwise the user logic around Subchannel creation will
likely to race with handleSubchannelState().
Will log a warning if LoadBalancer.Helper.createSubChannel() is called
outside of the SynchronizationContext.
Adds SynchronizationContext.throwIfNotInThisSynchronizationContext()
to facilitate this warning. It can also be used by LoadBalancer
implementations to make it a requirement.
LoadBalancerProvider is the interface that extends LoadBalancer.Factory. LoadBalancerRegistry is the one that loads the providers through service loader, and allows users to access providers through their names.
pick_first and round_robin balancer factories, which are experimental public API are now deprecated. Their providers are internal, as they are accessible by policy name.
AutoConfiguredLoadBalancerFactory is modified to access implementations purely by their names, thus hard-coded class names are no longer needed, and it can support arbitrary policy selected by service config.
The trailing dot denotes the hostname to be absolute. It is fine to
leave, but removing it makes the authority match the more common form
and hopefully reduces confusion.
This happens to works around SNI failures caused when using gRPC-LB,
since SNI prohibits the trailing dot. However, that is not the reason
for this change as we have to support users directly providing a
hostname with the trailing dot anyway (and doing so is not hard).
See #4912
Swapping MetadataApplier to an abstract class is not ABI-safe for
callers. So I revert back to the previous interface definition and
introduce a CallCredentials2.MetadataApplier which is an abstract class.
Once everyone is on CallCredentials2 then we can swap it to an abstract
class again.
Fixes#5002
Provides a `SynchronizationContext` for scheduling tasks, with and without delay, from LoadBalancer implementations. This absorbs and extends the internal utility `ChannelExecutor`. It supersedes `Helper.runSerialized()`, which is now deprecated.
# Motivation
I see multiple cases that schedule tasks with a delay while requiring the task to run in the "Channel Executor". There have been repeated work to wrap scheduled tasks and handle races between cancellation and task run (see the diff in `GrpclbState.java` for example). The LoadBalancer implementation (e.g., GrpclbLoadBalancer) also has to acquire the `ScheduledExecutorService` from somewhere and release it upon shutdown.
The upcoming HealthCheckLoadBalancer (#4932), which would use back-off policy to retry health-checking streams, would have to do all the things above. At this point I think we need to provide something that combines `runSerialized()` with a scheduled executor with the same synchronization guarantees.
# Design details
`SynchronizationContext` is a similar to `ScheduledExecutorService` but tailored for use in `LoadBalancer` and potentially other cases outside of `LoadBalancer`. It offers task queuing and serialization and delayed scheduling. It guarantees non-reentrancy and happens-before among tasks. It owns no thread, but run tasks on caller's or caller-provided threads.
All channel-level state mutations and callback methods on `LoadBalancer` are done in a SynchronizationContext, which was previously referred to as "Channel Executor".
`SynchronizationContext.schedule()` returns a `ScheduledHandle` for status checking and cancellation. `ScheduedFuture` from `SchedulingExecutorService.schedule()` is too broad for our use cases (e.g., the blocking `get()` should never be used).
`SynchronizationContext.schedule()` requires a `ScheduledExecutorService`, which is now available through `Helper.getScheduledExecutorService()`. LoadBalancers don't need to worry about where to get `SchedulingExecutorService` any more.
# Alternatives
Alternatively, we could keep `Helper.runSerialized()` and add something like `Helper.runSerialiezdWithDelay()`, but having them on their own interface allows clean fake implementation by `FakeClock` for test, and allows other components (potentially `InternalSubchannel` for reconnection backoff) to use it too.
Instead of asking caller of `schedule()` to provide the `ScheduledExecutorService`, we considered having SynchronizationContext take a `ScheduledExecutorService` at construction. It would be inconvenient for LoadBalancer implementations that don't use `schedule()`, as they would be forced to provide a fake `ScheduledExecutorService` (which is cumbersome).
Instead of making `SynchronizationContext` a (semi-)concrete class, we considered making it an pure abstract class. However, we found it nontrivial to implement `execute()` correctly with the non-reentrancy guarantee.
This reverts commit ef8a84421d.
Firebase is not yet ready to migrate to the new API. Will try again once we made the release and migrated them to CallCredentials2.
This will allow enabling Error Prone on JDK 10+ (after
updating the net.ltgt.errorprone plugin), and is also a
prerequisite to that plugin update.
Also remove net.ltgt.apt plugin, as Gradle has native
support for annotationProcessor.
This change is mainly to fix a test, but it also is an implementation of the proposal here: https://github.com/grpc/proposal/pull/79
In short:
* Do not do SRV or TXT lookups when the target name is `localhost`. This can be overriden by a system property
* Do not do SRV or TXT lookups when the target name is an IPv6 or IPv4 address. This _cannot_ be overriden. The constructed domains for these queries would themselves not be valid. (e.g. _grpclb._tcp.192.168.0.1)
* Speeds up initial connection when communicating over local host, since it is extremely uncommon that such a connection would need gRPCLB or SRV records
I expect to remove the system property after a release if no one asks about it.
This is a rename of the pre-existing Netty builder method, so aliases
were added to the Netty builders.
Fixes#4050. This API was a minor rename to the pre-existing Netty API,
so has already undergone API review and thus is not ExperimentalApi.
There is a known issue that causes DNS lookup issue when network
siwtchover on android. This issue is tracked separately in #4962.
This change simply disables DNS cache to avoid the issue on Android.
Those overrides are kept for backward compatibility and convenience
for callers. Documentation already says implementations should not
override them. Making them final reduces confusion around which
override should be verified in tests and be overridden in forwarding
classes, thus prevents bugs caused by such confusion.
Returns a Channel that allows a LoadBalancer to make auxiliary RPCs on already-established application connections. We need this to implement client-side health-checking (#4932)
See comments on the API for its semantics.
Notable changes:
- Transports are modified to use InUseStateAggregator so that they can exclude RPCs made on Subchannel.asChannel() when reporting in-use state for idle mode.
- OobChannel shares the same Executor as Subchannel.asChannel(). Because the latter is not a ManagedChannel and doesn't have life-cycle, thus can't determine when to return the Executor to a pool, the Executor is now returned only when ManagedChannelImpl is terminated.
This change does 3 main things (in 3 commits):
1. Refactor the resolution runnable to be testable
2. Add Finer level logging to aid in debugging
3. Check that there are addresses before passing them to ManagedChannelImpl.
This is the first step of smoothly changing the CallCredentials API.
Security level and authority are parameters required to be passed to
applyRequestMetadata(). This change wraps them, along with
MethodDescriptor and the transport attributes to RequestInfo, which is
more clear to the implementers.
ATTR_SECURITY_LEVEL is moved to the internal GrpcAttributes and
annotated as TransportAttr, because transports are required to set it,
but no user is actually reading them from
{Client,Server}Call.getAttributes().
ATTR_AUTHORITY is removed, because no transport is overriding it.
All involved interfaces are changed to abstract classes, as this will
make further API changes smoother.
The CallCredentials name is stabilized, thus we first introduce
CallCredentials2, ask CallCredentials implementations to migrate to
it, while GRPC accepting both at the same time, then replace
CallCredentials with CallCredentials2.
This will be used by LoadBalancer plugins that delegates to another,
which is what the new request routing (go/grpc-request-routing-design)
requires. This will also be used to wrap LoadBalancers to add
client-side health-checking functionality.
* doc: organize Attributes Keys with annotations.
Keys are annotated with the following annotations:
1. Grpc.TransportAttr: transport attributes returned by
{Client,Server}Call.getAttributes().
2. NameResolver.ResolutionResultAttr: attributes passed as the
argument of NameResolver.Listener.onAddresses() and
LoadBalancer.handleResolvedAddressGroups()
3. EquivalentAddressGroup.Attr: attributes from
EquivalentAddressGroups.
* Expand the usage of annotations to Attributes variables.
Log using new proto definition
- Remove io.grpc.BinaryLog.CallId because a call ID is now an AtomicLong
- Add the concept of "always included" and "never included" metadata
keys. This is needed because grpc-status-details-bin is already
logged in the binlog msg, and we will log grpc-trace-bin for the
census info.
- unit tests are effectively rewritten
* Remove redundant SubchannelPicker refreshes in RoundRobinLoadBalancer
- Ensure active subchannel list and round-robin index is only
regenerated/refreshed when it changes
- Make it so that Subchannels exist in subchannels map iff their state
!= SHUTDOWN
- Add EmptyPicker class since logic for this case is disjoint from the
non-empty case
* remove explicit initialization of boolean ready field
per @carl-mastrangelo's review comment
* minor restructuring to make logic clearer; more explanatory comments
* move some checks inside updateBalancingState method for clarity
* store current state and picker in RRLB, only update when new one is diff
* some more simplification/refactoring; improve test coverage
- remove now redundant check in handleSubchannelState
- collapse getAggregatedState() and getAggregatedError() into
handleBalancingState()
- have both pickers extend new RoundRobinPicker, move
areEquivalentPickers() logic into RoundRobinPicker.isEquivalentTo()
- extend unit tests to cover some additional cases
* Address latest review comments from @zhangkun83
- Use explicit check for non-empty list instead of assert
- Change EmptyPicker.status to be non-nullable
- Further test coverage improvement including explicit picker comparison
tests
* use EMPTY_OK instead of Status.OK for initial empty picker
The exact census span behavior wrt gRPC is not yet defined, so let's
punt on tight integration.
It may be fine to log grpc-trace-bin on server side because it is a
key visible to the application.
This is an API used to coordinate across packages and must live in
`io.grpc`.
Prepending `Internal` makes it easier to detect and hide this class
from public visibility when using certain build tools.
fixes#4796
* Make the list of providers an immutable List
* Make obvious that the list is statically initialized
* Add documentation for when methods were added.
* Use RuntimeException, rather than IllegalStateException.
There seem to be some users converting from StatusRuntimeException
to StatusException using the following paradigm:
sre.getStatus().toException(sre.getTrailers())
Since there isn't a viable alternative, revert the warning.
grpc-netty is still really useful, but for most users who aren't doing
anything advanced using grpc-netty-shaded is much safer from a
dependency basis.
grpc-netty-shaded has seen more usage and has shown itself to be stable
and reduce the number of conflicts due to Netty versions.
RoundRobinLoadBalancerFactory creates a new Picker instance every time the set of provided address groups changes or the connection state of subchannels associated with existing address groups changes. In certain scenarios, such as deployment/replacement of the target service cluster, this can lead to high churn of Picker objects. Given that each new Picker's subchannel index is initialized to zero, in these scenarios requests can end up getting disproportionately routed through subchannels (and hence server nodes) which are earlier in the list of address groups.
At Netflix we have measured that some service nodes end up taking 3-4x the load that of other nodes during deployment.
This commit randomizes the start index of the RoundRobinLoadBalancerFactory.Picker which eliminates this behavior.
Prevent multiple effective close calls either by successful completion
of a cancel or complete notification, or through successive exceptions
handled within a single call.
DoNotMock was removed from error_prone_annotations in 2.1.3, because
there was no enforcement mechanism (which is in google/error-prone#572).
Guava and Trust also depend on error_prone_annotations and are beginning
to use newer versions, so our usage of DoNotMock is causing diamond
dependency problems. This allows us to update to 2.2.0.
The annotations were useful internally; we're solving that in cl/205294089.
Most of the changes are changing the signature of newClientTransport.
Since this is annoying, I choose to introduce a ClientTransportOptions
object to avoid the churn in the future.
With ClientTransportOptions in place, there's only a few lines necessary
of plumbing for the Attributes: add the field to ClientTransportOptions
and populate it in InternalSubchannel. There are no consumers of the
field in this commit.
This avoids the needs to flatten to EAGs for cases like PickFirst,
making the Attributes in EAGs able to be used in communication with
core. See #4302 for some discussion on the topic.
Instead of failing after a a missing A/AAAA record, this change
makes the resolver keep going and try out SRV records too. This
is needed for use with ALTS, and is part of the gRPCLB spec.
This change also moved the JNDI code to a separate, reflectively
loaded file. This makes it easy to exclude the file and not worry
about the missing class references on Android. Additionally, if
javax.naming might be available on Android, this allows it to be
loaded. A key side effect of this is that DnsNameResolver is
smaller, and more cleanly tested.
The motivation here is in some cases we log the remote-addr that is set in the gRPC call attributes, and have to special case this type to support inprocess servers.
Eric Anderson
ZHANG Dapeng
zpencer
Kun Zhang
Carl Mastrangelo
Thomas Broyer
Jihun Cho
Jesse Wilson
Nick Hill
Eric Gribkoff
Tom Leach
George Gensure
Yang Song
Marco Ferrer
Doug Lawrie
Grant Oakley