apple/foundationdb
1
0
Fork 0
mirror of https://github.com/apple/foundationdb.git synced 2025-05-15 02:18:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

only skip confirm epoch live if CAUSAL_READ_RISKY is enabled

Browse Source 
time checked on the proxy should be less than the time waited by the master to account for clock speed differences
setting REQUIRED_MIN_RECOVERY_DURATION and ENFORCED_MIN_RECOVERY_DURATION to 0 will go back to the old behavior
This commit is contained in:
Evan Tschannen 2019-07-12 17:58:16 -07:00
parent a63969afb3
commit 02de53160d
15 changed files with 47 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
bindings/flow/tester/Tester.actor.cpp
View File

@ -1562,6 +1562,7 @@ struct UnitTestsFunc : InstructionFunc {
data->db->setDatabaseOption(FDBDatabaseOption::FDB_DB_OPTION_TRANSACTION_SIZE_LIMIT, Optional<StringRef>(StringRef((const uint8_t*)&sizeLimit, 8)));
data->db->setDatabaseOption(FDBDatabaseOption::FDB_DB_OPTION_TRANSACTION_RETRY_LIMIT, Optional<StringRef>(StringRef((const uint8_t*)&retryLimit, 8)));
data->db->setDatabaseOption(FDBDatabaseOption::FDB_DB_OPTION_TRANSACTION_RETRY_LIMIT, Optional<StringRef>(StringRef((const uint8_t*)&noRetryLimit, 8)));
data->db->setDatabaseOption(FDBDatabaseOption::FDB_DB_OPTION_CAUSAL_READ_RISKY);
data->db->setDatabaseOption(FDBDatabaseOption::FDB_DB_OPTION_SNAPSHOT_RYW_ENABLE);
data->db->setDatabaseOption(FDBDatabaseOption::FDB_DB_OPTION_SNAPSHOT_RYW_DISABLE);

1
bindings/python/tests/tester.py
View File

@ -138,6 +138,7 @@ def test_db_options(db):
db.options.set_transaction_size_limit(100000)
db.options.set_transaction_retry_limit(10)
db.options.set_transaction_retry_limit(-1)
db.options.set_causal_read_risky()
db.options.set_snapshot_ryw_enable()
db.options.set_snapshot_ryw_disable()

1
bindings/ruby/tests/tester.rb
View File

@ -462,6 +462,7 @@ class Tester
@db.options.set_transaction_size_limit(100000)
@db.options.set_transaction_retry_limit(10)
@db.options.set_transaction_retry_limit(-1)
@db.options.set_causal_read_risky()
@db.options.set_snapshot_ryw_enable()
@db.options.set_snapshot_ryw_disable()

2
documentation/sphinx/source/api-c.rst
View File

@ -49,6 +49,8 @@
.. |max-retry-delay-database-option| replace:: FIXME
.. |transaction-size-limit-database-option| replace:: FIXME
.. |timeout-database-option| replace:: FIXME
.. |causal-read-risky-transaction-option| replace:: FIXME
.. |causal-read-risky-database-option| replace:: FIXME
.. include:: api-common.rst.inc

6
documentation/sphinx/source/api-common.rst.inc
View File

@ -318,6 +318,10 @@
Set the default maximum backoff delay incurred by each transaction in the call to |on-error-func| if the error is retryable. This is equivalent to calling |max-retry-delay-transaction-option| on each transaction created by this database.
.. |option-db-causal-read-risky-blurb| replace::
Transactions do not require the strict causal consistency guarantee that FoundationDB provides by default. The read version will be committed, and usually will be the latest committed, but might not be the latest committed in the event of a simultaneous fault and misbehaving clock. Enabling this option is equivalent to calling |causal-read-risky-transaction-option| on each transaction created by this database.
.. |option-db-snapshot-ryw-enable-blurb| replace::
If this option has been set an equal or more times with this database than the disable option, snapshot reads *will* see the effects of prior writes in the same transaction. Enabling this option is equivalent to calling |snapshot-ryw-enable-transaction-option| on each transaction created by this database.
@ -351,7 +355,7 @@
.. |option-causal-read-risky-blurb| replace::
This transaction does not require the strict causal consistency guarantee that FoundationDB provides by default. The read version of the transaction will be a committed version, and usually will be the latest committed, but it might be an older version in the event of a fault or network partition.
This transaction does not require the strict causal consistency guarantee that FoundationDB provides by default. The read version will be committed, and usually will be the latest committed, but might not be the latest committed in the event of a simultaneous fault and misbehaving clock. One can set this for all transactions by calling |causal-read-risky-database-option|.
.. |option-causal-write-risky-blurb| replace::

6
documentation/sphinx/source/api-python.rst
View File

@ -25,6 +25,7 @@
.. |timeout-database-option| replace:: :func:`Database.options.set_transaction_timeout`
.. |max-retry-delay-database-option| replace:: :func:`Database.options.set_transaction_max_retry_delay`
.. |transaction-size-limit-database-option| replace:: :func:`Database.options.set_transaction_size_limit`
.. |causal-read-risky-database-option| replace:: :func:`Database.options.set_causal_read_risky`
.. |snapshot-ryw-enable-database-option| replace:: :func:`Database.options.set_snapshot_ryw_enable`
.. |snapshot-ryw-disable-database-option| replace:: :func:`Database.options.set_snapshot_ryw_disable`
.. |future-type-string| replace:: a :ref:`future <api-python-future>`
@ -35,6 +36,7 @@
.. |size-limit-transaction-option| replace:: :func:`Transaction.options.set_size_limit`
.. |snapshot-ryw-enable-transaction-option| replace:: :func:`Transaction.options.set_snapshot_ryw_enable`
.. |snapshot-ryw-disable-transaction-option| replace:: :func:`Transaction.options.set_snapshot_ryw_disable`
.. |causal-read-risky-transaction-option| replace:: :func:`Transaction.options.set_causal_read_risky`
.. |lazy-iterator-object| replace:: generator
.. |key-meth| replace:: :meth:`Subspace.key`
.. |directory-subspace| replace:: :ref:`DirectorySubspace <api-python-directory-subspace>`
@ -384,6 +386,10 @@ Database options
|option-db-tr-size-limit-blurb|
.. method:: Database.options.set_causal_read_risky()
|option-db-causal-read-risky-blurb|
.. method:: Database.options.set_snapshot_ryw_enable()
|option-db-snapshot-ryw-enable-blurb|

6
documentation/sphinx/source/api-ruby.rst
View File

@ -23,6 +23,7 @@
.. |timeout-database-option| replace:: :meth:`Database.options.set_transaction_timeout`
.. |max-retry-delay-database-option| replace:: :meth:`Database.options.set_transaction_max_retry_delay`
.. |transaction-size-limit-database-option| replace:: :func:`Database.options.set_transaction_size_limit`
.. |causal-read-risky-database-option| replace:: :meth:`Database.options.set_causal_read_risky`
.. |snapshot-ryw-enable-database-option| replace:: :meth:`Database.options.set_snapshot_ryw_enable`
.. |snapshot-ryw-disable-database-option| replace:: :meth:`Database.options.set_snapshot_ryw_disable`
.. |future-type-string| replace:: a :class:`Future`
@ -33,6 +34,7 @@
.. |size-limit-transaction-option| replace:: :meth:`Transaction.options.set_size_limit`
.. |snapshot-ryw-enable-transaction-option| replace:: :meth:`Transaction.options.set_snapshot_ryw_enable`
.. |snapshot-ryw-disable-transaction-option| replace:: :meth:`Transaction.options.set_snapshot_ryw_disable`
.. |causal-read-risky-transaction-option| replace:: :meth:`Transaction.options.set_causal_read_risky`
.. |lazy-iterator-object| replace:: :class:`Enumerator`
.. |key-meth| replace:: :meth:`Subspace.key`
.. |directory-subspace| replace:: :class:`DirectorySubspace`
@ -380,6 +382,10 @@ Database options
|option-db-tr-size-limit-blurb|
.. method:: Database.options.set_causal_read_risky() -> nil
|option-db-causal-read-risky-blurb|
.. method:: Database.options.set_snapshot_ryw_enable() -> nil
|option-db-snapshot-ryw-enable-blurb|

2
documentation/sphinx/source/data-modeling.rst
View File

@ -51,6 +51,8 @@
.. |max-retry-delay-database-option| replace:: FIXME
.. |transaction-size-limit-database-option| replace:: FIXME
.. |timeout-database-option| replace:: FIXME
.. |casual-read-risky-database-option| replace:: FIXME
.. |causal-read-risky-transaction-option| replace:: FIXME
.. include:: api-common.rst.inc

2
documentation/sphinx/source/developer-guide.rst
View File

@ -51,6 +51,8 @@
.. |max-retry-delay-database-option| replace:: FIXME
.. |transaction-size-limit-database-option| replace:: FIXME
.. |timeout-database-option| replace:: FIXME
.. |casual-read-risky-database-option| replace:: FIXME
.. |causal-read-risky-transaction-option| replace:: FIXME
.. include:: api-common.rst.inc

2
fdbclient/NativeAPI.actor.cpp
View File

@ -760,7 +760,7 @@ void DatabaseContext::setOption( FDBDatabaseOptions::Option option, Optional<Str
if (defaultFor >= 0) {
ASSERT(FDBTransactionOptions::optionInfo.find((FDBTransactionOptions::Option)defaultFor) !=
FDBTransactionOptions::optionInfo.end());
transactionDefaults.addOption((FDBTransactionOptions::Option)option, value.castTo<Standalone<StringRef>>());
transactionDefaults.addOption((FDBTransactionOptions::Option)defaultFor, value.castTo<Standalone<StringRef>>());
}
else {
switch(option) {

5
fdbclient/vexillographer/fdb.options
View File

@ -162,6 +162,9 @@ description is not currently required but encouraged.
paramType="Int" paramDescription="value in bytes"
description="Set the maximum transaction size in bytes. This sets the ``size_limit`` option on each transaction created by this database. See the transaction option description for more information."
defaultFor="503"/>
<Option name="transaction_casual_read_risky" code="504"
description="The read version will be committed, and usually will be the latest committed, but might not be the latest committed in the event of a simultaneous fault and misbehaving clock."
defaultFor="20"/>
<!-- The snapshot RYW options act like defaults for the equivalent transaction options, but database defaults cannot have cumulative effects from multiple calls.
Thus, we don't use the defaultFor annotation on these options. -->
<Option name="snapshot_ryw_enable" code="26"
@ -174,7 +177,7 @@ description is not currently required but encouraged.
<Option name="causal_write_risky" code="10"
description="The transaction, if not self-conflicting, may be committed a second time after commit succeeds, in the event of a fault"/>
<Option name="causal_read_risky" code="20"
description="The read version will be committed, and usually will be the latest committed, but might not be the latest committed in the event of a fault or partition"/>
description="The read version will be committed, and usually will be the latest committed, but might not be the latest committed in the event of a simultaneous fault and misbehaving clock."/>
<Option name="causal_read_disable" code="21" />
<Option name="next_write_no_write_conflict_range" code="30"
description="The next write performed on this transaction will not generate a write conflict range. As a result, other transactions which read the key(s) being modified by the next write will not conflict with this transaction. Care needs to be taken when using this option on a transaction that is shared between multiple threads. When setting this option, write conflict ranges will be disabled on the next write operation, regardless of what thread it is on." />

5
fdbserver/Knobs.cpp
View File

@ -285,10 +285,11 @@ ServerKnobs::ServerKnobs(bool randomize, ClientKnobs* clientKnobs) {
init( MAX_TXS_POP_VERSION_HISTORY, 1e5 );
init( PROXY_FORWARD_DELAY, 10.0 );
init( MAX_FORWARD_MESSAGES, 1e6 ); if( randomize && BUGGIFY ) MAX_FORWARD_MESSAGES = 10;
init( MIN_CONFIRM_INTERVAL, 0.05 );
bool shortRecoveryDuration = randomize && BUGGIFY;
init( MIN_RECOVERY_DURATION, 0.1 ); if( shortRecoveryDuration ) MIN_RECOVERY_DURATION = 0.01;
init( MAX_COMMIT_LATENCY, 0.09 ); if( shortRecoveryDuration ) MAX_COMMIT_LATENCY = 0.009;
init( ENFORCED_MIN_RECOVERY_DURATION, 0.085 ); if( shortRecoveryDuration ) ENFORCED_MIN_RECOVERY_DURATION = 0.01;
init( REQUIRED_MIN_RECOVERY_DURATION, 0.080 ); if( shortRecoveryDuration ) REQUIRED_MIN_RECOVERY_DURATION = 0.01;
// Master Server
// masterCommitter() in the master server will allow lower priority tasks (e.g. DataDistibution)

5
fdbserver/Knobs.h
View File

@ -230,8 +230,9 @@ public:
int MAX_TXS_POP_VERSION_HISTORY;
double PROXY_FORWARD_DELAY;
int MAX_FORWARD_MESSAGES;
double MIN_RECOVERY_DURATION;
double MAX_COMMIT_LATENCY;
double MIN_CONFIRM_INTERVAL;
double ENFORCED_MIN_RECOVERY_DURATION;
double REQUIRED_MIN_RECOVERY_DURATION;
// Master Server
double COMMIT_SLEEP_TIME;

14
fdbserver/MasterProxyServer.actor.cpp
View File

@ -267,7 +267,7 @@ struct ProxyCommitData {
getConsistentReadVersion(getConsistentReadVersion), commit(commit), lastCoalesceTime(0),
localCommitBatchesStarted(0), locked(false), commitBatchInterval(SERVER_KNOBS->COMMIT_TRANSACTION_BATCH_INTERVAL_MIN),
firstProxy(firstProxy), cx(openDBOnServer(db, TaskPriority::DefaultEndpoint, true, true)), db(db),
singleKeyMutationEvent(LiteralStringRef("SingleKeyMutation")), commitBatchesMemBytesCount(0), lastTxsPop(0), lastStartCommit(0), lastCommitLatency(SERVER_KNOBS->MIN_RECOVERY_DURATION), lastCommitTime(0)
singleKeyMutationEvent(LiteralStringRef("SingleKeyMutation")), commitBatchesMemBytesCount(0), lastTxsPop(0), lastStartCommit(0), lastCommitLatency(SERVER_KNOBS->REQUIRED_MIN_RECOVERY_DURATION), lastCommitTime(0)
{}
};
@ -1144,8 +1144,10 @@ ACTOR Future<GetReadVersionReply> getLiveCommittedVersion(ProxyCommitData* commi
for (auto const& p : *otherProxies)
proxyVersions.push_back(brokenPromiseToNever(p.getRawCommittedVersion.getReply(GetRawCommittedVersionRequest(debugID), TaskPriority::TLogConfirmRunningReply)));
if (!(flags&GetReadVersionRequest::FLAG_CAUSAL_READ_RISKY) && (now() - SERVER_KNOBS->MIN_RECOVERY_DURATION) > commitData->lastCommitTime.get()) {
wait(commitData->lastCommitTime.whenAtLeast(now() - SERVER_KNOBS->MIN_RECOVERY_DURATION));
if (!(flags&GetReadVersionRequest::FLAG_CAUSAL_READ_RISKY)) {
wait(commitData->logSystem->confirmEpochLive(debugID));
} else if (SERVER_KNOBS->REQUIRED_MIN_RECOVERY_DURATION > 0 && now() - SERVER_KNOBS->REQUIRED_MIN_RECOVERY_DURATION > commitData->lastCommitTime.get()) {
wait(commitData->lastCommitTime.whenAtLeast(now() - SERVER_KNOBS->REQUIRED_MIN_RECOVERY_DURATION));
}
if (debugID.present())
@ -1534,7 +1536,7 @@ ACTOR Future<Void> updateLastCommit(ProxyCommitData* self) {
ACTOR Future<Void> lastCommitUpdater(ProxyCommitData* self, PromiseStream<Future<Void>> addActor) {
loop {
double interval = (SERVER_KNOBS->MIN_RECOVERY_DURATION - std::min(SERVER_KNOBS->MAX_COMMIT_LATENCY, self->lastCommitLatency))/2;
double interval = std::max(SERVER_KNOBS->MIN_CONFIRM_INTERVAL, (SERVER_KNOBS->REQUIRED_MIN_RECOVERY_DURATION - self->lastCommitLatency)/2.0);
double elapsed = now()-self->lastStartCommit;
if(elapsed < interval) {
wait( delay(interval + 0.0001 - elapsed) );
@ -1608,7 +1610,9 @@ ACTOR Future<Void> masterProxyServerCore(
// wait for txnStateStore recovery
wait(success(commitData.txnStateStore->readValue(StringRef())));
addActor.send(lastCommitUpdater(&commitData, addActor));
if(SERVER_KNOBS->REQUIRED_MIN_RECOVERY_DURATION > 0) {
addActor.send(lastCommitUpdater(&commitData, addActor));
}
int commitBatchByteLimit =
(int)std::min<double>(SERVER_KNOBS->COMMIT_TRANSACTION_BATCH_BYTES_MAX,

2
fdbserver/masterserver.actor.cpp
View File

@ -1258,7 +1258,7 @@ ACTOR Future<Void> masterCore( Reference<MasterData> self ) {
if(oldLogSystem) {
logChanges = triggerUpdates(self, oldLogSystem);
if(!minRecoveryDuration.isValid()) {
minRecoveryDuration = delay(SERVER_KNOBS->MIN_RECOVERY_DURATION);
minRecoveryDuration = delay(SERVER_KNOBS->ENFORCED_MIN_RECOVERY_DURATION);
}
}