Kinetic is a key-value storage system. A Kinetic Device (e.g. a Kinetic Drive or a traditional server running the Java Reference Implementation) stores key-value objects. Kinetic Client applications can communicate with a Kinetic Device by sending messages over a network using TCP. Each individual message is called a “Kinetic Protocol Data Unit” (Kinetic PDU) and represents an individual request or response. For example, a Kinetic Client may send a message requesting the value associated with a particular key to a Kinetic Device. The device would respond with a message containing the value.
This document describes the structure of Protocol Buffer messages in detail. It is important to have a familiarity with the Protocol Buffer data interchange format (https://code.google.com/p/protobuf/). Where data types are specified with respect to fields in protobuf
messages, the Scalar Value Types documented here: https://developers.google.com/protocol-buffers/docs/proto will be used.
- Overview
- Kinetic Protocol Data Unit Structure - Protobuf Structure
- Access Control
- Operation Details
- Overview
- Cross-Cutting Concerns
- No Op
- Modify Value Operations
- Cross-Cutting Concerns
- PUT
- Delete
- Flush
- [Synchronization Modes and Behavior] (#user-content-synchronization-modes-and-behavior)
- Read Operations
- Setup
- Administration
- Peer to Peer
- Batch Operation
Table of Contents generated with DocToc
A Kinetic Protocol Data Unit is composed of a Protocol Buffer (protobuf
) message, containing operation metadata & key-value metadata, and the value. It is important to note that the value is not encoded in the protobuf
message; it is a separate top-level component of the Kinetic PDU.
Specifically, a Kinetic PDU is structured as follows:
Offset | Type | Length | Description |
---|---|---|---|
0 | Byte | 1 Byte | Version prefix: currently the character ‘F’, denoting the beginning of the message. (The character ‘F’, the Hex value 46). |
1 | 4 Byte Big Endian Integer | 4 Bytes | The number of bytes in the protobuf message (the maximum length for protobuf messages is 1024*1024 bytes). |
5 | 4 Byte Big Endian Integer | 4 Bytes | The number of bytes in the value (the maximum length for values is 1024*1024 bytes). |
9 | Bytes | <= 1024*1024 Bytes | The protobuf message. |
9 + length of protobuf message |
Bytes | <= 1024*1024 Bytes | The value. |
Within a Kinetic PDU, the protobuf
message encodes the specifics of the requested operation (or response). At a high level, each protobuf
message contains:
- A required AuthType that indicates the authentication type of the Kinetic PDU message.
- An optional HMACauth that contains userId and HMAC of the byte representation of the Command.
- An optional PINauth that contains byte representation of the pin.
- a commandBytes field that contains the
protobuf
byte representation of the Command.
Each Command contains a:
- Header, containing metadata about the message such as type (e.g. GET, GET_RESPONSE, PUT, PUT_RESPONSE, etc)
- Body, containing operation-specific information, such as key-value information for PUT or key range information for GETKEYRANGE.
- Status, containing information about whether an associated operation succeeded or failed (and why).
The message structure for each operation will be described in depth in the following sections.
The Kinetic Protocol supports restricting the operations a requester (identity) can perform by way of Access Control Lists (ACLs). They are structured as follows:
message ACL {
// The same identity specified in the header of messages
optional int64 identity = 1;
// This is the identity's HMAC Key. This is a shared secret between the
// client and the device, used to sign requests.
optional bytes key = 2;
// This is the algorithm used for performing the HMAC for messages for
// this identity.
// The supported values are: HmacSHA1.
optional HMACAlgorithm hmacAlgorithm = 3;
// Scope is the core of an ACL, an identity can have several.
// See below.
repeated Scope scope = 4;
// Scopes grant a set of permissions to the identity associated
// with the ACL. Scopess can further restrict which situations
// those permissions apply to by using the offset, value,
// and TlsRequired fields
message Scope {
// Offset and value are optional and should be used to restrict
// which keys the Scope applies to. If offset and value are
// specified, the permission will only apply to keys that match
// the value at the given offset. This is analogous to a substring
// match in many languages, where the key in question is the target.
optional int64 offset = 1;
optional bytes value = 2;
// The Permission being granted.
// There can be many, there must be at least one.
repeated Permission permission = 3;
// Optional boolean, defaults to false.
// When set to true, this scope only applies to SSL connections.
// Even if an identity has an ACL with a scope containing a specific
// permission, if that permission belongs to a scope for which
// TlsRequired is true and the identity makes a non-ssl request,
// Kinetic will behave as if the identity does not have that
// permission.
optional bool TlsRequired = 4;
}
// These are the permissions that can be included in a scope
enum Permission {
INVALID = -1; // place holder for backward compatibility
READ = 0; // can read key/values
WRITE = 1; // can write key/values
DELETE = 2; // can delete key/values
RANGE = 3; // can do a range
SETUP = 4; // can set up a device
P2POP = 5; // can do a peer to peer operation
GETLOG = 7; // can get log
SECURITY = 8; // can set up the security permission of the device
}
// Currently only one valid HMAC algorithm is supported
enum HMACAlgorithm {
// Added to allow additional HmacAlgorithms without breaking
// backward compatibility.
Unknown = 0;
// this is the default
HmacSHA1 = 1;
}
}
See the Security section below for details on setting ACLs.
In this section we'll give some concrete examples of how ACLs can be used.
###Client 1
Suppose client 1 has an ACL like so:
ACL {
identity: 1
key: "a3b38c37298f7f01a377518dae81dd99655b2be8129c3b2c6357b7e779064159"
HMACAlgorithm: HmacSHA1
// There can be multiple scopes, we'll show that in these examples by
// repeated scope objects like this
scope {
permission: READ
}
scope {
offset: 0
value: "foo"
permission: WRITE
}
}
Client 1 would be able to GET
any object in the store, but only PUT
keys that start with "foo".
###Client 2
Suppose client 2 has an ACL like so:
ACL {
identity: 2
key: "13010b8d8acdbe6abc005840aad1dc5dedb4345e681ed4e3c4645d891241d6b2"
HMACAlgorithm: HmacSHA1
scope {
permission: SECURITY
TlsRequired: true
}
}
Client 2 would be able to create new identities and set ACLs (using the Security operation) but only over SSL connections. Client 2 would not be able to read or write any keys on the device (though they could reset their own ACL to allow such activity).
This section describes the protobuf
message structure for each operation supported by the Kinetic protocol. There are many fields that may be set on all requests, to simplify this document those will be documented once in the Cross-Cutting Concerns section. Within each logical grouping of operations (read value, modify value, etc) there are additional common fields. We will begin each sub-section with a description of common fields.
There are many fields in the protobuf
message which can be specified on many operations. Instead of repeating the documentation for those fields for each call, we will show them here.
Request Message
Message {
// required AuthType
// Every message must be one of the defined enum auth types (HMACAUTH|PINAUTH|UNSOLICITEDSTATUS).
authType: ...
// optional HMACauth
// Normal messages uses this auth type
hmacAuth: ...
// optional PINauth
// for Pin based operations. These include device unlock and
// device erase
pinAuth: ...
// required bytes
// the embedded message providing the request (for HMACauth) and
// the response (for all auth types).
// the protocol buffer Command message is encoded/decoded to/from the commandBytes bytes
commandBytes: ...
// The Message Type determines how the the message is to be processed.
enum AuthType {
// This is for normal traffic. Check the HMAC of the command and
// if correct, process the command.
HMACAUTH: ...
// device unlock and ISE command. These must come over the TLS connection.
// If they do not, close the connection. If it is over
// the TLS connection, execute the pin operation.
PINAUTH: ...
// In the event that the device is going to close the connection, an
// unsolicited status will be returned first.
UNSOLICITEDSTATUS = 3;
}
// This is for normal message to the device
// and for responses. These are allowed once the
// device is unlocked. The HMAC provides for
// authenticity, Integrity and to enforce roles.
message HMACauth {
// optional int64
// The "identity" identifies the requester and the key and algorithm to
// be used for hmac.
identity: ...
// optional bytes
// The HMAC of this message used to verify integrity.
// The HMAC is taken of the byte-representaiton of the Command message of this
// protobuf message. An identity-specific shared secret is used to compute the HMAC.
// The Kinetic Device must have the key associated with the identity in
// this HMACauth message.
// For example, in pseudocode where a computeHMAC function exists which takes
// a value and an algorithm:
// hmac = computeHMAC(message.command.toBytes(), identityHMACAlgorithm)
hmac: ...
}
// Pin based authentication for Pin operations.
message PINauth {
// optional bytes
// The pin necessary to make the operations valid
pin: ...
}
}
Command {
header {
// Optional int64, default value is 0
// The version number of this cluster definition. If this is not equal to
// the value on the device, the request is rejected and will return a
// `VERSION_FAILURE` `statusCode` in the `Status` message.
clusterVersion: ...
// Required int64
// The identity associated with this request. See the ACL discussion above.
// The Kinetic Device will use this identity value to lookup the
// HMAC key (shared secret) to verify the HMAC.
identity: ...
// Required int64
// A unique number for this connection between the source and target.
// On the first request to the drive, this should be the time of day in
// seconds since 1970. The drive can change this number and the client must
// continue to use the new number and the number must remain constant
// during the session
connectionID: ...
// Required int64
// Sequence is a monotonically increasing number for each request in a TCP
// connection.
sequence: ...
// Required MessageType
// The message type identifies which sort of operation this is.
// See the MessageType enum in the protobuf definition for all potential
// values.
// Note that the *_RESPONSE message types are reserved for messages from
// the Kinetic Device to the client (i.e. responses).
messageType: ...
}
body {
// Omitted in this cross-cutting documentation section
}
}
Response Message
Message {
// required AuthType
// see request message above.
authType: ...
// optional HMACauth
// see request message above
hmacAuth: ...
// required bytes
// the protocol buffer Command message is encoded/decoded to/from the commandBytes bytes
// see request message above
commandBytes: ...
}
Command {
header {
// Required int64.
// In a response message, ackSequence will be the same as the
// sequence value set in the request message.
// The client can use this to map async responses to their
// associated requests.
// This is important because operations within a connection may be reorderd.
ackSequence: ...
// In a response, messageType corresponds to the requested messageType.
// For instance, requests with a PUT messageType will receive a response
// with a PUT_REPONSE messageType.
messageType: ...
}
body {
// Omitted in this cross-cutting documentation section
}
status {
// Every response from the Kinetic Device will specify a code indicating
// whether the request was successful, or the specific error case
// encountered. The full list of codes is specified by the
// Status.StatusCode enum.
code: SUCCESS
}
}
###Error Cases###
When an error occurs on the Kinetic Device, the response message includes a status
with a code
. These codes are enumerated in the StatusCode
enum in the protocol definition. They will be discussed here in more detail.
INTERNAL_ERROR
indicates that the Kinetic Device experiences a malfunction. (Currently this code is returned in certain cases that don't indicate a drive malfunction, these will be updated.)HMAC_FAILURE
indicates that the HMAC of the request is incorrect or missing. This will also be returned when an unknown identity is set in the header, since the device cannot verify an HMAC for an unknown identity.NOT_AUTHORIZED
indicates the attemped operation could not be completed because the identity set in the header did not have authorization. This may mean that the identity does not have the required Permission in any Scope in the ACL, or it may indicate that the Scope containing that Permission does not apply (due to offset & index or tls rules).VERSION_FAILURE
indicates that theclusterVersion
of the Kinetic Device does not match theclusterVersion
set in the header of the requesting message.NOT_FOUND
indicates that the requested key was not found in the Kinetic Device's data store.VERSION_MISMATCH
indicates that thePUT
orDELETE
operation failed because thedbVersion
passed in theKeyValue
object does not match the store's version. Pasingforce: true
in theKeyValue
object ignores the mismatch and completes the operation.NO_SPACE
indicates that the drive is full. There are background processes which may free space, so this error may occur once, and not on subsequent tries even though no data has been explicitly removed. Similarly, executing a delete may not immediately free space, so aPUT
which fails with this error may not immediately succeed even after aDELETE
which should free space.NO_SUCH_HMAC_ALGORITHM
indicates that thehmacAlgorithm
field in theSecurity
message was invalid.INVALID_REQUEST
indicates that the request is not valid. Subsequent attempts with the same request will return the same code. Examples: GET does not specify keyValue message, GETKEYRANGE operation does not specify startKey, etc.NOT_ATTEMPTED
indicates that a P2P operation was received but was not even attempted due to some other error halting execution early.REMOTE_CONNECTION_ERROR
indicates that a P2P operation was attempted but could not be completed.NESTED_OPERATION_ERRORS
indicates that a P2P request completed but that an operation (possibly nested) failed.EXPIRED
indicates that an operation did not complete in the alotted time.DEVICE_LOCKED
indicates that the device is currently locked and can not valida the hmac. The connection is terminated after the device sent the response message.DEVICE_ALREADY_UNLOCKED
indicates that the device was already unlocked. The validity of the pin was NOT checked. The connection remains open.CONNECTION_TERMINATED
indicates that the connection is being terminated. Details as to why are set in the message string.INVALID_BATCH
If an error occurs before END BATCH is received, the device sends an unsolicited status message with status code set to INVALID BATCH and closed the connection. If an error occurs after END BATCH is received, an INVALID BATCH status message is returned and the connection is remained open. If the drive is locked or received ISE command before an END BATCH is received, the drive will return an unsolicited status message (in LOCK state) and close the connection.
A number of error codes are defined in the protocol file but not currently used:
HEADER_REQUIRED
SERVICE_BUSY
DATA_ERROR
PERM_DATA_ERROR
It is possible that an error will occur that will prevent the Kinetic Device from returning a protobuf
message with a status code. These are some situations:
- Invalid Kinetic PDU: If the Kinetic PDU is not formed as described above, the TCP connection will be closed abruptly. This includes the case that a value or protobuf message exceeds the size limitations.
- Invalid Protobuf: If the
protobuf
message cannot be decoded because it is not well formed, the TCP connection will be closed abruptly.
The NOOP
operation can be used as a quick test of whether the Kinetic Device is running and available. If the Kinetic Device is running, this operation will always return succeed.
Request Message
message {
// See above for descriptions of these fields
authType: HMACAUTH
hmacAuth {
identity: ...
hmac: "..."
}
commandBytes: "..."
}
// human readable commandBytes (decoded from commandBytes)
command {
header {
// See above for descriptions of these fields
clusterVersion: ...
connectionID: ...
sequence: ...
// messageType should be NOOP
messageType: NOOP
}
}
Response Message
message {
authType: HMACAUTH
hmacAuth {
identity: ...
hmac: "..."
}
commandBytes: "..."
}
// human readable commandBytes (decoded from commandBytes)
command {
header {
// See above
connectionID: ...
ackSequence: ...
// messageType should be NOOP_RESPONSE
messageType: NOOP_RESPONSE
}
status {
code: SUCCESS
}
}
Within the body
message of a value modification operation, many fields in the keyValue
apply to all operations.
command: {
...
body: {
keyValue {
// Required bytes
// The key for the value being set
key: "..."
// Required bytes
// Versions are set on objects to support optimistic locking.
// For operations that modify data, if the dbVersion sent in the
// request message does not match the version stored in the db, the
// request will fail.
dbVersion: "..."
// Required bytes
// Specifies what the next version of the data will be if this
// operation is successful.
newVersion: "..."
// Optional bool, default false
// Setting force to true ignores potential version mismatches
// and carries out the operation.
force: true
// Optional bytes
// The integrity value for the data. This value should be computed
// by the client application by applying the hash algorithm
// specified below to the value (and only to the value).
// The algorithm used should be specified in the algorithm field.
// The Kinetic Device will not do any processing on this value.
tag: "..."
// The algorithm used by the client to compute the tag.
// The allowed values are: SHA1, SHA2, SHA3, CRC32, CRC64
algorithm: ...
// Optional Synchronization enum value, defaults to WRITETHROUGH
// Allows client to specify if the data must be written to disk
// immediately, or can be written in the future.
//
// WRITETHROUGH: This request is made persistent before returning.
// This does not effect any other pending operations.
// WRITEBACK: They can be made persistent when the drive chooses,
// or when a subsequent FLUSH is give to the drive.
// FLUSH: All pending information that has not been written is
// pushed to the disk and the command that specifies
// FLUSH is written last and then returned. All WRITEBACK writes
// that have received ending status will be guaranteed to be
// written before the FLUSH operation is returned completed.
synchronization: ...
}
}
}
The PUT
operation sets the value and metadata for a given key. If a value already exists in the store for the given key, the client must pass a value for dbVersion
which matches the stored version for this key to overwrite the value metadata. This behavior can be overridden (so that the version is ignored and the value and metadata are always written) by setting forced
to true
in the KeyValue
option.
Request Message
The following request will add a key value pair to the store. Note that dbVersion
is not specified, this is allowed when adding (as opposed to updating) a value.
command {
// See top level cross cutting concerns for header details
header {
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// The messageType should be PUT
messageType: PUT
}
body {
keyValue {
// See write operation cross cutting concerns
newVersion: "..."
key: "..."
tag: "..."
algorithm: ...
synchronization: ...
}
}
}
Response Message When the key is successfully written, the device will respond with the following message:
command {
header {
// See above
ackSequence: ...
// The messageType should be PUT_RESPONSE
messageType: PUT_RESPONSE
}
body {
keyValue {
// Empty
}
}
status {
// A successful PUT will return SUCCESS
code: SUCCESS
}
}
Error Cases:
code = VERSION_MISMATCH
- For a PUT of a new key (insert, not update) specifying a dbVersion
- If the version doesn't match (should not occur for create)
code = NOT_AUTHORIZED
- If the identity doesn't have permission to put this value, in this case
status.statusMessage
will be "permission denied."
- If the identity doesn't have permission to put this value, in this case
- The connection will be closed without reply if the value is too long. (The result in a client library may be some sort of IO Error, depending on implementation).
The DELETE
operation removes the entry for a given key. It respects the same locking behavior around dbVersion
and force
as described in the previous sections.
Request Message
The following request will remove a key value pair to the store.
command {
// See top level cross cutting concerns for header details
header {
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// messageType should be DELETE
messageType: DELETE
}
body {
keyValue {
key: "..."
// See write operation cross cutting concerns
synchronization: ...
}
}
}
Response Message When the entry is successfully removed, the device will respond with the following message:
command {
// See top level cross cutting concerns for header details
header {
ackSequence: ...
// messageType should be DELETE_RESPONSE
messageType: DELETE_RESPONSE
}
body {
keyValue {
}
}
status {
// A successful DELETE will return SUCCESS
code: SUCCESS
}
}
There are many cases where a delete could fail with a properly functioning drive. The following status.code
values identify these cases:
code = VERSION_MISMATCH
The dbVersion in the request doesn't match the version stored in the device.code = NOT_FOUND
The key was not found in the data store.code = NOT_AUTHORIZED
The identity doesn't have permission to delete this value, in this casestatus.statusMessage
will be "permission denied."
The FLUSHALLDATA
operation flushes any outstanding PUTs or DELETEs on the device. For example, if the client PUT
many keys with synchronization=WRITEBACK
the data
would not be guaranteed to be persisted, so power cycling could result in lost data. When a FLUSHALLDATA
command returns, all previous operations with synchronization=WRITEBACK
on
this connection are guaranteed to be persisted. Data on separate connections is not guaranteed to be persisted, but may as an indirect consequence of this operation.
Request Message
The following request will flush the write cache.
command {
// See top level cross cutting concerns for header details
header {
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// messageType should be FLUSHALLDATA
messageType: FLUSHALLDATA
}
body {
}
}
Response Message When the cache is flushed, the device will return the following message:
command {
// See top level cross cutting concerns for header details
header {
ackSequence: ...
// messageType should be FLUSHALLDATA_RESPONSE
messageType: FLUSHALLDATA_RESPONSE
}
body {
}
status {
// A successful FLUSHALLDATA will return SUCCESS
code: SUCCESS
}
}
If no system or power failure occurred, the behavior for PUT/DELETE WRITETHROUGH mode and PUT/DELETE WRITEBACK mode are functionally equivalent except for the response time differences. That is, the database integrity is maintained for both WRITETHROUGH and WRITEBACK modes.
Read and modify K/V operations (such as PUT/GET/DELETE) sent from a client within a single threaded context on the same TCP/SSL connection SHOULD BE processed and responded with the received order.
Permissions No special permissions are required.
There are a number of operations which are designed to allow clients to read values from the Kinetic Device. They will be discussed in this section.
Within the body
message of a read value operation, many fields in the keyValue
message apply to all operations.
keyValue {
// Required bytes.
// The key identifying the value in the data store.
key: "..."
// Optional bool, defaults to false.
// If true, only metadata (not the full value) will be returned
// If false, metadata and value will be returned
metadataOnly: ...
}
The GET
operation is used to retrieve the value and metadata for a given key.
Request Message
command {
header {
// See above for descriptions of these fields
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// The mesageType should be GET
messageType: GET
}
body {
keyValue {
// See above
key: "..."
}
}
}
Response Message
A successful response will return the value in the top level Kinetic PDU, and will have a SUCCESS
status:
command {
header {
// See above
ackSequence: ...
// messageType should be GET_RESPONSE
messageType: GET_RESPONSE
}
body {
keyValue {
// These fields are documented above
key: "..."
dbVersion: "..."
tag: ""
algorithm: SHA2
}
}
status {
code: SUCCESS
}
}
There are many cases where a read could fail with a properly functioning drive. The following status.code
values identify these cases:
NOT_FOUND
The key does not exist in the data store (the Kinetic PDU will have a zero-length value component).NOT_AUTHORIZED
The identity doesn't have permission to put this value, in this casestatus.statusMessage
will be "permission denied."
The GETVERSION
operation provdes the current store version for a given key.
Request Message
command {
header {
// These fields are documented above
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// messageType should be GETVERSION
messageType: GETVERSION
}
body {
keyValue {
// Required. See above.
key: "..."
}
}
}
Response Message
command {
header {
// This field is documented above
ackSequence: ...
// messageType should be GETVERSION_RESPONSE
messageType: GETVERSION_RESPONSE
}
body {
keyValue {
// The dbVersion is the only entry in the keyValue object that will
// be returned by the server
dbVersion: "..."
}
}
status {
code: SUCCESS
}
}
Error Cases:
code = NOT_FOUND
The key does not exist in the data store (the Kinetic PDU will have a zero-length value component).code = NOT_AUTHORIZED
The requester doesn't have permission to put this value, in this casestatus.statusMessage
will be "permission denied."
The GETNEXT
operation takes a key and returns the value for the next key in the sorted set of keys. Keys are sorted lexicographically by their byte representation.
Request Message
command {
header {
// See above for descriptions of these fields
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// messageType should be GETNEXT
messageType: GETNEXT
}
body {
keyValue {
// A key is required. Note that this is different from GET in that you
// will not get the value for this key, but the value for the subsequent
// key in the ordering.
key: "..."
}
}
}
Response Message
A successful response will return the value in the top level Kinetic PDU, and will have a SUCCESS
status:
command {
header {
// See above for descriptions of this field
ackSequence: ...
// messageType should be GETNEXT_RESPONSE
messageType: GETNEXT_RESPONSE
}
body {
keyValue {
// This is the key for the value that is being returned
// This will be different from the key passed in the request
key: "..."
dbVersion: "..."
tag: ""
algorithm: SHA2
}
}
status {
// If the operation does not succeed, a different code will be specified.
// See below.
code: SUCCESS
}
}
Error Cases:
code = NOT_FOUND
- There is no key in the store that is sorted after the given key.
- This can occur if the given key is the last key in the store, of if the key given is not included in the store but would be sorted after the last key.
code = NOT_AUTHORIZED
The identity does not have read permission on the key that would be returned.
Edge Cases:
- If a
key
is provided which is not found in the store, the service will find the first key which would be sorted after the given key. For example, if the store has keyskey0
andkey2
and the client sends a request forGETNEXT
ofkey1
, the device will return the value forkey2
. - Note that if the identity does not have permission to read the key passed in the
GETNEXT
request, but they do have permission to read the key that would be returned, the request should succeed.
The GETPREVIOUS
operation takes a key and returns the value for the previous key in the sorted set of keys. Keys are sorted lexicographically by their byte representation.
Request Message
command {
header {
// See above for descriptions of these fields
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// messageType should be GETPREVIOUS
messageType: GETPREVIOUS
}
body {
keyValue {
// A key is required. Note that this is different from GET in that you
// will not get the value for this key, but the value for the subsequent
// key in the ordering.
key: "..."
}
}
}
Response Message
A successful response will return the value in the top level Kinetic PDU, and will have a SUCCESS
status:
command {
header {
// See above for descriptions of this field
ackSequence: ...
// messageType should be GETPREVIOUS_RESPONSE
messageType: GETPREVIOUS_RESPONSE
}
body {
keyValue {
// This is the key for the value that is being returned
// This will be different from the key passed in the request
key: "..."
// These fields are documented above
dbVersion: "..."
tag: "..."
algorithm: ...
}
}
status {
// If the operation does not succeed, a different code will be specified.
// See below.
code: SUCCESS
}
}
Error Cases:
code = NOT_FOUND
- There is no key in the store that is sorted brefore the given key.
- This can occur if the given key is the first key in the store, of if the key given is not included in the store but would be sorted before the first key.
code = NOT_AUTHORIZED
:- If the identity does not have read permission on the key that would be returned.
Edge Cases:
- If a
key
is provided which is not found in the store, the service will find the first key which would be sorted before the given key. For example, if the store has keyskey0
andkey2
and the client sends a request forGETPREVIOUS
ofkey1
, the device will return the value forkey0
. - Note that if the identity does not have permission to read the key passed in the
GETNEXT
request, but they do have permission to read the key that would be returned, the request should succeed.
The GETKEYRANGE
operation takes a start and end key and returns all keys between those in the sorted set of keys. This operation can be configured so that the range is either inclusive or exclusive of the start and end keys, the range can be reversed, and the requester can cap the number of keys returned.
Note that this operation does not fetch associated values, or other metadata. It only returns the keys themselves, which can be used for other operations.
Request Message
command {
header {
// See above for descriptions of these fields
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// messageType should be GETKEYRANGE
messageType: GETKEYRANGE
}
body {
// The range message must be populated
range {
// Required bytes, the beginning of the requested range
startKey: "..."
// Optional bool, defaults to false
// True indicates that the start key should be included in the returned
// range
startKeyInclusive: ...
// Required bytes, the end of the requested range
endKey: "..."
// Optional bool, defaults to false
// True indicates that the end key should be included in the returned
// range
endKeyInclusive: ...
// Required int32, must be greater than 0
// The maximum number of keys returned, in sorted order
maxReturned: ...
// Optional bool, defaults to false
// If true, the key range will be returned in reverse order, starting at
// endKey and moving back to startKey. For instance
// if the search is startKey="j", endKey="k", maxReturned=2,
// reverse=true and the keys "k0", "k1", "k2" exist
// the system will return "k2" and "k1" in that order.
reverse: ....
}
}
}
Response Message
command {
header {
ackSequence: ...
messageType: GETKEYRANGE_RESPONSE
}
body {
// The range message is populated with up to maxReturned keys.
// If no keys are found in the range then the range message will be omitted
// and the status code will be SUCCESS
range {
key: "..."
key: "..."
...
key: "..."
}
}
status {
code: SUCCESS
}
}
Error Cases:
code = INVALID_REQUEST
- The
maxReturned
exceeded the limit, thestatus.statusMessage
will beKey limit exceeded.
- The
Edge Cases:
- If neither
startKey
orendKey
are found in the store, any keys that would be sorted between them will be returned. - If the given keys are out of order (e.g.
startKey
is sorted afterendKey
), then no keys will be returned.
Permissions
This operation should return the first contiguous block of keys for which the requesting identity has the RANGE
permission on an applicable scope. This means that not necessarily each key in the requested range for which the identity has this permission will be returned. For instance, consider a store that contains k0
, k1
, k2
, k4
, and k5
, where the requesting identity has the RANGE
permission on scopes which aply to k0
, k1
, k4
, and k5
but notably does not have RANGE
permission on any scope which applies to k2
. Then if that identity requests a GETKEYRANGE
with startKey=k0
(inclusive), endKey=k5
(inclusive) the Kinetic Device will return k0
and k1
. When it reaches k2
, for which it does not have a RANGE
permission, it will stop the operation.
The SETUP
operation can be used to set the device's clusterVersion
and pin
, to perform an "Instant Secure Erase", or to download new firmware on the device. As these operations are quite different, we'll discuss them separately in this section. The Kinetic Device will only allow one of these operations per message (though syntactically several could be combined).
Request Message
command {
header {
// Important: this should be the current cluster version. This operation is
// intended to change the clusterVersion, but the current clusterVersion
// must be specified here.
clusterVersion: ...
// See top level cross cutting concerns for header details
identity: ...
connectionID: ...
sequence: ...
// The messageType should be SETUP
messageType: SETUP
}
body {
setup {
// Required int64, needed to update the cluster version
// (otherwise request will be treated as a different Setup operation)
// This is the clusterVersion being set on the device.
newClusterVersion: 1
}
}
}
Response Message
command {
header {
ackSequence: ...
// The messageType should be SETUP_RESPONSE
messageType: SETUP_RESPONSE
}
status {
code: SUCCESS
}
}
This operation should be used to erase all stored data from the device.
Request Message
command {
header {
// See top level cross cutting concerns for header details
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// The messageType should be SETUP
messageType: SETUP
}
body {
setup {
// Required bool, defaults to false if omitted.
// Must be true for this request to be treated as an ISE.
instantSecureErase: true
}
}
}
Response Message
command {
header {
ackSequence: ...
// The messageType should be SETUP_RESPONSE
messageType: SETUP_RESPONSE
}
status {
code: SUCCESS
}
}
This operation should be used load new firmware on the device.
Request Message
command {
header {
// See top level cross cutting concerns for header details
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// The messageType should be SETUP
messageType: SETUP
}
body {
setup {
// Required bool, must be present and true to indicate that this is
// a firmware download operation.
// Indicates that the value (in the Kinetic PDU) will contain the firmware
firmwareDownload: true
}
}
}
The value field in the Kinetic PDU (describe above) will contain the firmware payload.
Response Message
command {
header {
ackSequence: ...
// The messageType should be SETUP_RESPONSE
messageType: SETUP_RESPONSE
}
status {
code: SUCCESS
}
}
The security operation allows administrators to specify ACLs, granting access to specific operations. Some semantics of the Security operation are noteworthy:
- A
identity
has one ACL, and an ACL only applies to oneidentity
. They have a one-to-one relationship. - An ACL list cannot be updated, only set. Each request to SECURITY with a well-formed security body will overwrite the existing setup.
- To make a Secuirty operation (set ACLs) the requesting identity must have an applicable scope with a SECURITY permission.
To set the ACL for a identity (or many identities), a request like the following could be sent. See the Access Control section above for further explanation of the ACL message.
Request Message
command {
header {
// See top level cross cutting concerns for header details
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// messageType should be SECURITY
messageType: SECURITY
}
body {
// The security message must be present and contain at least one acl
// message. Multiple are allowed but only one can be specified per identity.
// Note that security message overwrites the stored ACL list entirely,
// no updating is supported.
security {
acl {
// Required int64, the identity this ACL applies to
identity: 1
// Required bytes, the identity's HMAC key, a shared secret
key: "...."
// Required HMACAlgorithm, the algorithm used to compute the HMAC for
// this identity
hmacAlgorithm: ...
// The scope message has at least one permission, in this example there
// are many
scope {
permission: READ
permission: WRITE
permission: DELETE
permission: RANGE
permission: SETUP
permission: P2POP
permission: GETLOG
permission: SECURITY
}
}
// This ACL shows that multiple scopes can be set for a identity in one
// ACL message
acl {
identity: 2
key: "..."
hmacAlgorithm: ...
// This simple scope allows identity 2 to read all keys
scope {
permission: READ
}
// This scope gives identity 2 the ability to write keys if and only if
// "test" is a substring of key starting at offset 3. For example, with
// this scope identity 2 could PUT keys: "xyztest1", "001test2", etc
// but could not put keys: "somethingElse", "test123", "1234test"
scope {
offset: 3
value: "test"
permission: WRITE
}
}
// More ACLs for additional identities may be specified in the
// same security message...
acl {
identity: 3
key: "..."
hmacAlgorithm: ...
scope {
permission: WRITE
}
}
acl {
identity: 4
key: "..."
hmacAlgorithm: ...
scope {
permission: DELETE
}
}
}
}
Response Message
command {
header {
ackSequence: ...
// messageType should be SECURITY_RESPONSE
messageType: SECURITY_RESPONSE
}
status {
// If successful, code will be SUCCESS
code: SUCCESS
}
}
Error Cases:
code=NOT_AUTHORIZED
if the requesting identity does not have theSECURITY
permission for an applicable scope.code=NO_SUCH_HMAC_ALGORITHM
if anacl
message has anhmaclAlgorithm
value which is invalid.code=INTERNAL_ERROR
(in the future, this code will be changed)- if an offset is provided which is less than zero
- if there are no permissions provided in a scope
- if one of the permissions provided is invalid (e.g. Permission.INVALID)
The GETLOG
operation gives the client access to log information. The request message must include at least one type and can have many types. The supported types are:
UTILIZATIONS
TEMPERATURES
CAPACITIES
CONFIGURATION
STATISTICS
MESSAGES
LIMITS
Below we will show the message structure used to request all types in a single GETLOG
request.
Request Message
command {
header {
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
// The messageType should be GETLOG
messageType: GETLOG
}
body {
// The body should contain a getLog message, which must have
// at least one value for type. Multiple are allowed.
// Here all types are requested.
getLog {
type: CAPACITIES
type: CONFIGURATION
type: MESSAGES
type: STATISTICS
type: TEMPERATURES
type: UTILIZATIONS
}
}
}
Respose Message
command {
header {
ackSequence: ...
// messageType should be GETLOG_RESPONSE
messageType: GETLOG_RESPONSE
}
body {
getLog {
// Each type requested is provided in the response
type: CAPACITIES
type: CONFIGURATION
type: MESSAGES
type: STATISTICS
type: TEMPERATURES
type: UTILIZATIONS
type: LIMITS
// Many utilization messages may be returned
utilization {
// Required string, the name of the rescource being reported
// For example: HDA, ENO, CPU...
name: "..."
// Required float, the value for this resource's utilization.
// value will be between 0.00 and 1.00.
value: 0.2
}
utilization {
name: "...""
value: ...
}
...
// Many temperature messages may be returned
temperature {
// Required string, the name of the resource being reported
name: "..."
// Required float, the current temperature in degrees celcius
current: 39.0
// Required float, the current temperature in degrees celcius
minimum: 5.0
// Required float, the current temperature in degrees celcius
maximum: 100.0
// Required float, the current temperature in degrees celcius
target: 25.0
}
// Only one configuration message will be included
configuration {
// string, the vendor of the Kinetic Device.
vendor: "..."
// string, the model of the Kinetic Device
model: "..."
// bytes, the serial number of the Kinetic Device
serialNumber: "..."
// string, the version of the kinetic software running on the device
version: "..."
// Multiple interface messages will appear, one per network interface
// that the Kinetic Device.
interface {
name: "..."
MAC: "..."
ipv4Address: "..."
ipv6Address: "..."
}
interface {
name: "..."
ipv4Address: "..."
ipv6Address: "..."
}
// int32, the port where the kinetic service is running
port: ...
// int32, the port where the kinetic service is running over SSL
tlsPort: ...
// string, he date this version of the kinetic service was compiled
compilationDate: "..."
// string, a checksum of the source code
sourceHash: "..."
}
// There should be one statistics message per messageType (GET, PUT, etc)
// The statistics messages aggregate statistics for each messageType.
statistics {
// Required MessageType, which messageType these statistics apply to
messageType: PUT
// Required sint64, how many times this messageType has been received
count: ...
// Required sint64, the sum length of all the value portion of the
// Kinetic PDU messages sent since starting the Kinetic Device
bytes: ...
}
...
statistics {
messageType: GET
count: ...
bytes: ...
}
// Only one capacity message will be included
capacity {
// uint64
nominalCapacityInBytes: ...
// float
portionFull: ...
}
// bytes representing recent Kinetic Device log messages
messages: "..."
// limits that the device will enforce
limits {
maxKeySize = ...
maxValueSize = ...
maxVersionSize = ...
maxTagSize = ...
maxConnections = ...
maxOutstandingReadRequests = ...
maxOutstandingWriteRequests = ...
maxMessageSize = ...
maxKeyRangeCount = ...
}
}
}
status {
code: SUCCESS
}
}
The PEER2PEERPUSH
operation allows a client to instruct a Kinetic Device to copy a set of keys (and associated value and metadata) to another Kinetic Device. Peer To Peer operations can be nested, so a client could tell device A to copy certain keys to device B, and then have device B copy a set of keys to device C, and so on.
Request Message
command {
header {
clusterVersion: ...
identity: ...
connectionID: ...
sequence: ...
messageType: PEER2PEERPUSH
}
body {
p2pOperation {
peer {
// Required string, the network address of the peer
hostname: "..."
// Required int32, the port on which the peer is running the Kinetic service
port: ...
// Optional boolean, defaults to false.
// Currently SSL is not supported so this must be false.
tls: ...
}
operation {
// Required bytes, the key to copy from the source peer.
key: ""
// Optional bytes, the
version: "..."
// Optional bool, defaults to false
// If true, force write ignoring version
force: ...
// Optional bytes, the key to use in the destination peer.
newKey: "..."
// This is a nested Peer To Peer Push operation. The recursive structure
// allows arbitrarily deep (up to the message size cap) nesting of
// p2p operations.
p2pop {
// Like the top-level p2pOperation, this specifies a peer and
// a set of operations
peer {
hostname: "..."
port: ...
tls: false
}
operation {
key: "..."
}
// Multiple operations can be specified in one P2POperation
operation {
key: "..."
}
}
}
}
}
}
Response Message
command {
header {
ackSequence: ...
messageType: PEER2PEERPUSH_RESPONSE
}
body {
p2pOperation {
// See below for a description of error handling
allChildOperationsSucceeded: false,
operation {
key: "..."
newKey: "..."
force: ...
status {
code: SUCCESS
}
p2pop {
peer {
hostname: "..."
port: ...
tls: false
}
// See below for a description of error handling
allChildOperationsSucceeded: false,
operation {
key: "..."
status {
// Status messages can be nested. This is what it would be
// returned if an operation failed because the key was not found
code: NOT_FOUND
}
}
operation: {
key: "...",
status {
code: NESTED_OPERATION_ERRORS
}
}
}
}
}
}
status {
code: SUCCESS
}
}
Error Cases:
If the command does not start or is terminated early, the status will be reflect that error.
If the request completed but some operations encountered errors, the code will be NESTED_OPERATION_ERRORS
.
If all operations and nested P2P Operations within the top-level operation are successful, the Status.code
in the Command
message will be SUCCESS
.
For each P2POperation, if any of its nested operations fail, then it will have the flag allChildOperationsSucceeded
set to false. Otherwise, that flag will be set to true.
Any operation may fail for the same reason any PUT
could fail. Operations have their own Status
message to report these failures.
In addition to the failures observed by PUT
, Operations may experience:
NOT_ATTEMPTED
The top level request was aborted before this operation could be attempted, either due to timeouts or another error (e.g. an IO error).REMOTE_CONNECTION_ERROR
The operation was attempted, but an error prevented the operation from completing.
Batch Operation allows a group of K/V commands (PUT and DELETE) to perform all at once. The commands within a batch are committed to the persistent store if all commands can be committed or otherwise nothing is committed.
A batch operation is started with a START_BATCH command and ended with a END_BATCH command. The supported operations are PUT and DELETE commands within a batch.
START_BATCH and END_BATCH have the request-response style messaging pattern similar to most of the Kinetic commands, such as GET command.
All (PUT/DELETE) operations within a batch do not have response messages.
The following Batch message construct is included in the END_BATCH and END_BATCH_RESPONSE messages.
// This is included in the END_BATCH and END_BATCH_RESPONSE.
message Batch {
// set by the client library in END_BATCH request message.
// the total number of operations in the batch
optional int32 count = 1;
// set by the drive in END_BATCH_RESPONSE message.
// If a batch is committed successfully, all sequence Ids of those
// commands (PUT/DELETE) performed in the batch are
// added in the END_BATCH_RESPONSE message.
repeated int64 sequence = 2 [packed=true];
// This field is set by the drive if a batch commit failed.
// The first failed operation sequence in the batch is set in the value of this field.
optional int64 failedSequence = 3;
}
START_BATCH Request Message
command {
header {
clusterVersion: ...
connectionID: ...
sequence: ...
// The messageType should be START_BATCH
messageType: START_BATCH
batchID: ...
}
body {
}
}
START_BATCH Response Message
command {
header {
connectionID: ...
ackSequence: ...
// The messageType should be START_BATCH_RESPONSE
messageType: START_BATCH_RESPONSE
}
status {
code: SUCCESS
}
}
END_BATCH Request Message
Command {
header {
clusterVersion: ...
connectionID: ...
sequence: ...
messageType: END_BATCH
batchID: ...
}
body {
batch {
count: 2
}
}
}
END_BATCH Response Message
Command {
header {
connectionID: ...
ackSequence: ...
messageType: END_BATCH_RESPONSE
}
body {
batch {
// see above description in message Batch construct
sequence: ...
sequence: ...
}
}
status {
code: SUCCESS
}
}
Error Cases:
If an error is detected before received the END_BATCH command, such as received more than maximum allowed commands within a batch, the device sent an Unsolicited Status Message and closed the connection. The StatusCode is set to INVALID_REQUEST and the cause is set to the StatusMessage in the Unsolicited Status Message.
If an error is detected after received the END_BATCH command, such as encountered a version mismatch for a PUT command, the device sends a END_BATCH_RESPONSE message with status code set to INVALID_BATCH. The failed sequence number of the command that caused the failure is set in the failedSequence field of the END_BATCH_RESPONSE message.
If the device is LOCKed before an END BATCH is received, the device returns an Unsolicited Status Message (INVALID_REQUEST status code, Device Locked message) and the uncommitted batch is removed. If an END BATCH is received and the batch has started processing before LOCK request is received, the batch is processed before the device is LOCKed.
If an ISE command is received before an END BATCH is received, the device sends an Unsolicited Status Message and closes the connection. The uncommitted batch is removed.
If a batch command (ie, PUT, DELETE) is received but there is no associated START BATCH, the device sends an Unsolicited Status Message and closes the connection. The StatusCode is set to INVALID_REQUEST and the cause is set to the StatusMessage in the Unsolicited Status Message.
Example INVALID_BATCH response message
command {
header {
connectionID: ...
ackSequence: ...
messageType: END_BATCH_RESPONSE
}
body {
batch {
sequence: ...
sequence: ...
failedSequence: ...
}
}
status {
code: INVALID_BATCH
statusMessage: "Version mismatch"
}
}