Package org.apache.cassandra.service

Class ClientState

java.lang.Object
org.apache.cassandra.service.ClientState
public class ClientState extends Object
State related to a client connection.

  • Field Details

    • READABLE_SYSTEM_RESOURCES

      public static final com.google.common.collect.ImmutableSet<IResource> READABLE_SYSTEM_RESOURCES

    • PROTECTED_AUTH_RESOURCES

      public static final com.google.common.collect.ImmutableSet<IResource> PROTECTED_AUTH_RESOURCES

    • isInternal

      public final boolean isInternal

  • Constructor Details

  • Method Details

    • resetLastTimestamp

      public static void resetLastTimestamp(long nowMillis)

    • forInternalCalls

      public static ClientState forInternalCalls()
      Returns:
      a ClientState object for internal C* calls (not limited by any kind of auth).

    • forInternalCalls

      public static ClientState forInternalCalls(String keyspace)

    • forExternalCalls

      public static ClientState forExternalCalls(SocketAddress remoteAddress)
      Returns:
      a ClientState object for external clients (native protocol users).

    • cloneWithKeyspaceIfSet

      public ClientState cloneWithKeyspaceIfSet(String keyspace)
      Clone this ClientState object, but use the provided keyspace instead of the keyspace in this ClientState object.
      Returns:
      a new ClientState object if the keyspace argument is non-null. Otherwise do not clone and return this ClientState object.

    • getTimestamp

      public static long getTimestamp()
      This clock guarantees that updates for the same ClientState will be ordered in the sequence seen, even if multiple updates happen in the same millisecond.

    • getTimestampForPaxos

      public static long getTimestampForPaxos(long minUnixMicros)
      Returns a timestamp suitable for paxos given the timestamp of the last known commit (or in progress update).

      Paxos ensures that the timestamp it uses for commits respects the serial order of those commits. It does so by having each replica reject any proposal whose timestamp is not strictly greater than the last proposal it accepted. So in practice, which timestamp we use for a given proposal doesn't affect correctness but it does affect the chance of making progress (if we pick a timestamp lower than what has been proposed before, our new proposal will just get rejected).

      As during the prepared phase replica send us the last propose they accepted, a first option would be to take the maximum of those last accepted proposal timestamp plus 1 (and use a default value, say 0, if it's the first known proposal for the partition). This would most work (giving commits the timestamp 0, 1, 2, ... in the order they are commited) up to 2 important caveats: 1) it would give a very poor experience when Paxos and non-Paxos updates are mixed in the same partition, since paxos operations wouldn't be using microseconds timestamps. And while you shouldn't theoretically mix the 2 kind of operations, this would still be pretty unintuitive. And what if you started writing normal updates and realize later you should switch to Paxos to enforce a property you want? 2) this wouldn't actually be safe due to the expiration set on the Paxos state table.

      So instead, we initially chose to use the current time in microseconds as for normal update. Which works in general but mean that clock skew creates unavailability periods for Paxos updates (either a node has his clock in the past and he may no be able to get commit accepted until its clock catch up, or a node has his clock in the future and then once one of its commit his accepted, other nodes ones won't be until they catch up). This is ok for small clock skew (few ms) but can be pretty bad for large one.

      Hence our current solution: we mix both approaches. That is, we compare the timestamp of the last known accepted proposal and the local time. If the local time is greater, we use it, thus keeping paxos timestamps locked to the current time in general (making mixing Paxos and non-Paxos more friendly, and behaving correctly when the paxos state expire (as long as your maximum clock skew is lower than the Paxos state expiration time)). Otherwise (the local time is lower than the last proposal, meaning that this last proposal was done with a clock in the future compared to the local one), we use the last proposal timestamp plus 1, ensuring progress.

      Parameters:
      minUnixMicros - the max timestamp of the last proposal accepted by replica having responded to the prepare phase of the paxos round this is for. In practice, that's the minimum timestamp this method may return.
      Returns:
      a timestamp suitable for a Paxos proposal (using the reasoning described above). Note that contrarily to the getTimestamp() method, the return value is not guaranteed to be unique (nor monotonic) across calls since it can return it's argument (so if the same argument is passed multiple times, it may be returned multiple times). Note that we still ensure Paxos "ballot" are unique (for different proposal) by (securely) randomizing the non-timestamp part of the UUID.

    • getLastTimestampMicros

      public static long getLastTimestampMicros()

    • getDriverName

      public Optional<String> getDriverName()

    • getDriverVersion

      public Optional<String> getDriverVersion()

    • getClientOptions

      public Optional<Map<String,String>> getClientOptions()

    • setDriverName

      public void setDriverName(String driverName)

    • setDriverVersion

      public void setDriverVersion(String driverVersion)

    • setClientOptions

      public void setClientOptions(Map<String,String> clientOptions)

    • getCQLQueryHandler

      public static QueryHandler getCQLQueryHandler()

    • getRemoteAddress

      public InetSocketAddress getRemoteAddress()

    • getRawKeyspace

      public String getRawKeyspace()

    • getKeyspace

      public String getKeyspace() throws InvalidRequestException
      Throws:
      InvalidRequestException

    • setKeyspace

      public void setKeyspace(String ks)

    • login

      public void login(AuthenticatedUser user)
      Attempts to login the given user.

    • ensureAllKeyspacesPermission

      public void ensureAllKeyspacesPermission(Permission perm)

    • ensureKeyspacePermission

      public void ensureKeyspacePermission(String keyspace, Permission perm)

    • ensureAllTablesPermission

      public void ensureAllTablesPermission(String keyspace, Permission perm)

    • ensureTablePermission

      public void ensureTablePermission(String keyspace, String table, Permission perm)

    • ensureTablePermission

      public void ensureTablePermission(TableMetadataRef tableRef, Permission perm)

    • ensureTablePermission

      public void ensureTablePermission(TableMetadata table, Permission perm)

    • hasTablePermission

      public boolean hasTablePermission(TableMetadata table, Permission perm)

    • ensurePermission

      public void ensurePermission(Permission perm, IResource resource)

    • ensurePermission

      public void ensurePermission(Permission permission, Function function)

    • validateLogin

      public void validateLogin()

    • ensureNotAnonymous

      public void ensureNotAnonymous()

    • isOrdinaryUser

      public boolean isOrdinaryUser()
      Checks if this user is an ordinary user (not a super or system user).
      Returns:
      true if this user is an ordinary user, false otherwise.

    • isSuper

      public boolean isSuper()
      Checks if this user is a super user.

    • isSystem

      public boolean isSystem()
      Checks if the user is the system user.
      Returns:
      true if this user is the system user, false otherwise.

    • ensureIsSuperuser

      public void ensureIsSuperuser(String message)

    • warnAboutUseWithPreparedStatements

      public void warnAboutUseWithPreparedStatements(MD5Digest statementId, String preparedKeyspace)

    • getUser

      public AuthenticatedUser getUser()