Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

docs(spanner): document RetryPolicy interface #12037

Merged
merged 2 commits into from
Jul 6, 2023
Merged
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
288 changes: 261 additions & 27 deletions google/cloud/spanner/retry_policy.h
Original file line number Diff line number Diff line change
Expand Up @@ -59,33 +59,267 @@ GOOGLE_CLOUD_CPP_INLINE_NAMESPACE_END
namespace spanner {
GOOGLE_CLOUD_CPP_INLINE_NAMESPACE_BEGIN

/// The base class for retry policies.
using RetryPolicy = ::google::cloud::internal::TraitBasedRetryPolicy<
spanner_internal::SafeGrpcRetry>;

/// A retry policy that limits based on time.
using LimitedTimeRetryPolicy =
::google::cloud::internal::LimitedTimeRetryPolicy<
spanner_internal::SafeGrpcRetry>;

/// A retry policy that limits the number of times a request can fail.
using LimitedErrorCountRetryPolicy =
google::cloud::internal::LimitedErrorCountRetryPolicy<
spanner_internal::SafeGrpcRetry>;

/// The base class for transaction rerun policies.
using TransactionRerunPolicy = ::google::cloud::internal::TraitBasedRetryPolicy<
spanner_internal::SafeTransactionRerun>;

/// A transaction rerun policy that limits the duration of the rerun loop.
using LimitedTimeTransactionRerunPolicy =
google::cloud::internal::LimitedTimeRetryPolicy<
spanner_internal::SafeTransactionRerun>;

/// A transaction rerun policy that limits the number of failures.
using LimitedErrorCountTransactionRerunPolicy =
google::cloud::internal::LimitedErrorCountRetryPolicy<
spanner_internal::SafeTransactionRerun>;
/// The base class for the Spanner library retry policies.
class RetryPolicy : public google::cloud::RetryPolicy {
public:
/// Creates a new instance of the policy, reset to the initial state.
virtual std::unique_ptr<RetryPolicy> clone() const = 0;
};

/**
* A retry policy for `$connection_class_name$` based on counting errors.
devbww marked this conversation as resolved.
Show resolved Hide resolved
*
* This policy stops retrying if:
* - An RPC returns a non-transient error.
* - More than a prescribed number of transient failures is detected.
*
* In this class the following status codes are treated as transient errors:
* - [`kUnavailable`](@ref google::cloud::StatusCode)
* - [`kResourceExhausted`](@ref google::cloud::StatusCode)
* - [`kInternal`](@ref google::cloud::StatusCode) if the error message
* indicates a connection reset.
*/
class LimitedErrorCountRetryPolicy : public RetryPolicy {
public:
/**
* Create an instance that tolerates up to @p maximum_failures transient
* errors.
*
* @note Disable the retry loop by providing an instance of this policy with
* @p maximum_failures == 0.
*/
explicit LimitedErrorCountRetryPolicy(int maximum_failures)
: impl_(maximum_failures) {}

LimitedErrorCountRetryPolicy(LimitedErrorCountRetryPolicy&& rhs) noexcept
: LimitedErrorCountRetryPolicy(rhs.maximum_failures()) {}
LimitedErrorCountRetryPolicy(LimitedErrorCountRetryPolicy const& rhs) noexcept
: LimitedErrorCountRetryPolicy(rhs.maximum_failures()) {}

int maximum_failures() const { return impl_.maximum_failures(); }

bool OnFailure(Status const& s) override { return impl_.OnFailure(s); }
bool IsExhausted() const override { return impl_.IsExhausted(); }
bool IsPermanentFailure(Status const& s) const override {
return impl_.IsPermanentFailure(s);
}
std::unique_ptr<RetryPolicy> clone() const override {
return std::make_unique<LimitedErrorCountRetryPolicy>(
impl_.maximum_failures());
}

private:
google::cloud::internal::LimitedErrorCountRetryPolicy<
spanner_internal::SafeGrpcRetry>
impl_;
};

/**
* A retry policy for the Spanner library based on elapsed time.
*
* This policy stops retrying if:
* - An RPC returns a non-transient error.
* - The elapsed time in the retry loop exceeds a prescribed duration.
*
* The following status codes are treated as transient errors:
* - [`kUnavailable`](@ref google::cloud::StatusCode)
* - [`kResourceExhausted`](@ref google::cloud::StatusCode)
* - [`kInternal`](@ref google::cloud::StatusCode) if the error message
* indicates a connection reset.
*/
class LimitedTimeRetryPolicy : public RetryPolicy {
public:
/**
* Constructor given a `std::chrono::duration<>` object.
*
* @tparam DurationRep a placeholder to match the `Rep` tparam for
* @p maximum_duration's type. The semantics of this template parameter
* are documented in `std::chrono::duration<>`. In brief, the underlying
* arithmetic type used to store the number of ticks. For our purposes it
* is simply a formal parameter.
* @tparam DurationPeriod a placeholder to match the `Period` tparam for
* @p maximum_duration's type. The semantics of this template parameter
* are documented in `std::chrono::duration<>`. In brief, the length of
* the tick in seconds, expressed as a `std::ratio<>`. For our purposes it
* is simply a formal parameter.
* @param maximum_duration the maximum time allowed before the policy expires,
* while the application can express this time in any units they desire,
* the class truncates to milliseconds.
*
* @see https://en.cppreference.com/w/cpp/chrono/duration for more details
* about `std::chrono::duration`.
*/
template <typename DurationRep, typename DurationPeriod>
explicit LimitedTimeRetryPolicy(
std::chrono::duration<DurationRep, DurationPeriod> maximum_duration)
: impl_(maximum_duration) {}

LimitedTimeRetryPolicy(LimitedTimeRetryPolicy&& rhs) noexcept
: LimitedTimeRetryPolicy(rhs.maximum_duration()) {}
LimitedTimeRetryPolicy(LimitedTimeRetryPolicy const& rhs) noexcept
: LimitedTimeRetryPolicy(rhs.maximum_duration()) {}

std::chrono::milliseconds maximum_duration() const {
return impl_.maximum_duration();
}

std::unique_ptr<RetryPolicy> clone() const override {
return std::make_unique<LimitedTimeRetryPolicy>(impl_.maximum_duration());
}
bool OnFailure(Status const& s) override { return impl_.OnFailure(s); }
bool IsExhausted() const override { return impl_.IsExhausted(); }
bool IsPermanentFailure(Status const& s) const override {
return impl_.IsPermanentFailure(s);
}

// This is provided only for backwards compatibility.
using BaseType = RetryPolicy;

private:
google::cloud::internal::LimitedTimeRetryPolicy<
spanner_internal::SafeGrpcRetry>
impl_;
};

/**
* The base class for the Spanner library transaction rerun policies.
*
* The [`Client::Commit()`] functions (there are several overloads) consume a
* callable to create mutations or a list of mutations. `Commit()` creates a
* transaction and applies the mutations. If the transaction fails, an instance
* of this class is used to control whether the transaction will be attempted
* again.
*
* [`Client::Commit()`]: @ref google::cloud::spanner::Client::Commit()
*/
class TransactionRerunPolicy : public google::cloud::RetryPolicy {
public:
/// Creates a new instance of the policy, reset to the initial state.
virtual std::unique_ptr<TransactionRerunPolicy> clone() const = 0;
};

/**
* A transaction rerun policy based on counting errors.
*
* This policy stops running if:
* - An RPC returns a non-transient error.
* - More than a prescribed number of transient failures is detected.
*
* The following status codes are treated as transient errors:
* - [`kAborted`](@ref google::cloud::StatusCode)
* - [`kNotFound`](@ref google::cloud::StatusCode) when the error message
* indicates the missing (or expired) resource is the spanner session
* associated with the transaction.
*/
class LimitedErrorCountTransactionRerunPolicy : public TransactionRerunPolicy {
public:
/**
* Create an instance that tolerates up to @p maximum_failures transient
* errors.
*
* @note Disable the retry loop by providing an instance of this policy with
* @p maximum_failures == 0.
*/
explicit LimitedErrorCountTransactionRerunPolicy(int maximum_failures)
: impl_(maximum_failures) {}

LimitedErrorCountTransactionRerunPolicy(
LimitedErrorCountTransactionRerunPolicy&& rhs) noexcept
: LimitedErrorCountTransactionRerunPolicy(rhs.maximum_failures()) {}
LimitedErrorCountTransactionRerunPolicy(
LimitedErrorCountTransactionRerunPolicy const& rhs) noexcept
: LimitedErrorCountTransactionRerunPolicy(rhs.maximum_failures()) {}

int maximum_failures() const { return impl_.maximum_failures(); }

bool OnFailure(Status const& s) override { return impl_.OnFailure(s); }
bool IsExhausted() const override { return impl_.IsExhausted(); }
bool IsPermanentFailure(Status const& s) const override {
return impl_.IsPermanentFailure(s);
}
std::unique_ptr<TransactionRerunPolicy> clone() const override {
return std::make_unique<LimitedErrorCountTransactionRerunPolicy>(
impl_.maximum_failures());
}

// This is provided only for backwards compatibility.
using BaseType = TransactionRerunPolicy;

private:
google::cloud::internal::LimitedErrorCountRetryPolicy<
spanner_internal::SafeTransactionRerun>
impl_;
};

/**
* A transaction rerun policy based on elapsed time.
*
* This policy stops retrying if:
* - An RPC returns a non-transient error.
* - The elapsed time in the retry loop exceeds a prescribed duration.
*
* The following status codes are treated as transient errors:
* - [`kAborted`](@ref google::cloud::StatusCode)
* - [`kNotFound`](@ref google::cloud::StatusCode) when the error message
* indicates the missing (or expired) resource is the spanner session
* associated with the transaction.
*/
class LimitedTimeTransactionRerunPolicy : public TransactionRerunPolicy {
public:
/**
* Constructor given a `std::chrono::duration<>` object.
*
* @tparam DurationRep a placeholder to match the `Rep` tparam for
* @p maximum_duration's type. The semantics of this template parameter
* are documented in `std::chrono::duration<>`. In brief, the underlying
* arithmetic type used to store the number of ticks. For our purposes it
* is simply a formal parameter.
* @tparam DurationPeriod a placeholder to match the `Period` tparam for
* @p maximum_duration's type. The semantics of this template parameter
* are documented in `std::chrono::duration<>`. In brief, the length of
* the tick in seconds, expressed as a `std::ratio<>`. For our purposes it
* is simply a formal parameter.
* @param maximum_duration the maximum time allowed before the policy expires,
* while the application can express this time in any units they desire,
* the class truncates to milliseconds.
*
* @see https://en.cppreference.com/w/cpp/chrono/duration for more details
* about `std::chrono::duration`.
*/
template <typename DurationRep, typename DurationPeriod>
explicit LimitedTimeTransactionRerunPolicy(
std::chrono::duration<DurationRep, DurationPeriod> maximum_duration)
: impl_(maximum_duration) {}

LimitedTimeTransactionRerunPolicy(
LimitedTimeTransactionRerunPolicy&& rhs) noexcept
: LimitedTimeTransactionRerunPolicy(rhs.maximum_duration()) {}
LimitedTimeTransactionRerunPolicy(
LimitedTimeTransactionRerunPolicy const& rhs) noexcept
: LimitedTimeTransactionRerunPolicy(rhs.maximum_duration()) {}

std::chrono::milliseconds maximum_duration() const {
return impl_.maximum_duration();
}

bool OnFailure(Status const& s) override { return impl_.OnFailure(s); }
bool IsExhausted() const override { return impl_.IsExhausted(); }
bool IsPermanentFailure(Status const& s) const override {
return impl_.IsPermanentFailure(s);
}
std::unique_ptr<TransactionRerunPolicy> clone() const override {
return std::make_unique<LimitedTimeTransactionRerunPolicy>(
impl_.maximum_duration());
}

// Not very useful, but needed for backwards compatibility.
devbww marked this conversation as resolved.
Show resolved Hide resolved
using BaseType = RetryPolicy;

private:
google::cloud::internal::LimitedTimeRetryPolicy<
spanner_internal::SafeTransactionRerun>
impl_;
};

GOOGLE_CLOUD_CPP_INLINE_NAMESPACE_END
} // namespace spanner
Expand Down