diff options
Diffstat (limited to 'doc/peers-v2.0.txt')
| -rw-r--r-- | doc/peers-v2.0.txt | 288 |
1 files changed, 288 insertions, 0 deletions
diff --git a/doc/peers-v2.0.txt b/doc/peers-v2.0.txt new file mode 100644 index 0000000..5f3aa11 --- /dev/null +++ b/doc/peers-v2.0.txt @@ -0,0 +1,288 @@ + HAProxy's peers v2.0 protocol 08/18/2016 + +Author: Emeric Brun ebrun@haproxy.com + + +I) Encoded Integer and Bitfield. + + + 0 <= X < 240 : 1 byte (7.875 bits) [ XXXX XXXX ] + 240 <= X < 2288 : 2 bytes (11 bits) [ 1111 XXXX ] [ 0XXX XXXX ] + 2288 <= X < 264432 : 3 bytes (18 bits) [ 1111 XXXX ] [ 1XXX XXXX ] [ 0XXX XXXX ] + 264432 <= X < 33818864 : 4 bytes (25 bits) [ 1111 XXXX ] [ 1XXX XXXX ]*2 [ 0XXX XXXX ] + 33818864 <= X < 4328786160 : 5 bytes (32 bits) [ 1111 XXXX ] [ 1XXX XXXX ]*3 [ 0XXX XXXX ] + + + + +I) Handshake + +Each peer try to connect to each others, and each peer is listening +for a connect from other peers. + + +Client Server + Hello Message + ------------------------> + + Status Message + <------------------------ + +1) Hello Message + +Hello message is composed of 3 lines: + +<protocol> <version> +<remotepeerid> +<localpeerid> <processpid> <relativepid> + +protocol: current value is "HAProxyS" +version: current value is "2.0" +remotepeerid: is the name of the target peer as defined in the configuration peers section. +localpeerid: is the name of the local peer as defined on cmdline or using hostname. +processid: is the system process id of the local process. +relativepid: is the haproxy's relative pid (0 if nbproc == 1) + +2) Status Message + +Status message is a code followed by a LF. + +200: Handshake succeeded +300: Try again later +501: Protocol error +502: Bad version +503: Local peer name mismatch +504: Remote peer name mismatch + + +IV) Messages + +Messages: + +0 - - - - - - - 8 - - - - - - - 16 + Message Class| Message Type + +if Message Type >= 128 + +0 - - - - - - - 8 - - - - - - - 16 ..... + Message Class| Message Type | encoded data length | data + +Message Classes: +0: control +1: error +10: related to stick table updates +255: reserved + + +1) Control Messages Class + +Available message Types for class control: +0: resync request +1: resync finished +2: resync partial +3: resync confirm + + +a) Resync Request Message + +This message is used to request a full resync from a peer + +b) Resync Finished Message + +This message is used to signal remote peer that locally known updates have been pushed, and local peer was considered up to date. + +c) Resync Partial Message + +This message is used to signal remote peer that locally known updates have been pushed, and but the local peer is not considered up to date. + +d) Resync Confirm Message + +This message is an ack for Resync Partial or Finished Messages. + +It's allow the remote peer to go back to "on the fly" update process. + + +2) Messages Class + +Available message Types for this class are: +0: protocol error +1: size limit reached + +a) Protocol Message + +To signal that a protocol error occurred. Connection will be shutdown just after sending this message. + +b) Size Limit Error Message + +To signal that a message is outsized and can not be correctly handled. Connection will be broken. + + + +3) Stick Table Updates Messages Class + +Available message Types for this class are: +0: Entry update +1: Incremental entry update +2: table definition +3: table switch +4: updates ack message. + + +a) Update Message + +0 - - - - - - - 8 - - - - - - - 16 ..... + Message class | Message Type | encoded data length | data + + +data is composed like this + +0 - - - - - - - 32 ............................. +Local Update ID | Key value | data values .... + +Update ID in a 32bits identifier of the local update. + +Key value format depends of the table key type: + +- for keytype string + +0 ................................. +encoded string length | string value + +- for keytype integer +0 - - - - - - - - - - 32 +encoded integer value | + +- for other key type + +The value length is annonced in table definition message +0 .................... +value + + +b) Incremental Update Message + +Same format than update message except the Update ID is not present, the receiver should +consider that the update ID is an increment of 1 of the previous considered update message (partial or not) + + +c) Table Definition Message + +This message is used by the receiver to identify the stick table concerned by next update messages and +to know which data is pushed in these updates. + +0 - - - - - - - 8 - - - - - - - 16 ..... + Message class | Message Type | encoded data length | data + + +data is composed like this + +0 ................................................................... +Encoded Sender Table Id | Encoded Local Table Name Length | Table Name | Encoded Table Type | Encoded Table Keylen | Encoded Table Data Types Bitfield + + +Encoded Sender Table ID present a the id numerical ID affected to that table by the sender +It will be used by "Updates Aknowlegement Messages" and "Table Switch Messages". + +Encoded Local Table Name Length present the length to read the table name. + +"Table Name" is the shared identifier of the table (name of the current table in the configuration) +It permits the receiver to identify the concerned table. The receiver should keep in memory the matching +between the "Sender Table ID" to identify it directly in case of "Table Switch Message". + +Table Type present the numeric type of key used to store stick table entries: +integer + 2: signed integer + 4: IPv4 address + 5: IPv6 address + 6: string + 7: binary + +Table Keylen present the key length or max length in case of strings or binary (padded with 0). + +Data Types Bitfield present the types of data linearly pushed in next updates message (they will be linearly pushed in the update message) +Known types are +bit + 0: server id + 1: gpt0 + 2: gpc0 + 3: gpc0 rate + 4: connections counter + 5: connection rate + 6: number of current connections + 7: sessions counter + 8: session rate + 9: http requests counter + 10: http requests rate + 11: errors counter + 12: errors rate + 13: bytes in counter + 14: bytes in rate + 15: bytes out rate + 16: bytes out rate + 17: gpc1 + 18: gpc1 rate + +d) Table Switch Message + +After a Table Message Define, this message can be used by the receiver to identify the stick table concerned by next update messages. + +0 - - - - - - - 8 - - - - - - - 16 ..... + Message class | Message Type | encoded data length | data + + +data is composed like this + + +0 ..................... +encoded Sender Table Id + +c) Update Ack Message + +0 - - - - - - - 8 - - - - - - - 16 ..... + Message class | Message Type | encoded data length | data + +data is composed like this + +0 ....................... - - - - - - - - 32 +Encoded Remote Table Id | Update Id + + +Remote Table Id is the numeric identifier of the table on the remote side. +Update Id is the id of the last update locally committed. + +If a re-connection occurred, the sender should know they will have to restart the push of updates from this point. + +III) Initial full resync process. + + +a) Resync from local old process + +An old soft-stopped process will close all established sessions with remote peers and will try to connect to a new +local process to push all known ending with a Resync Finished Message or a Resync Partial Message (if it it does not consider itself as full updated). + +A new process will wait for a an incoming connection from a local process during 5 seconds. It will learn the updates from this +process until it receives a Resync Finished Message or a Resync Partial Message. If it receive a Resync Finished Message it will consider itself +as fully updated and stops to ask for resync. If it receive a Resync Partial Message it will wait once again for 5 seconds for an other incoming connection from a local process. +Same thing if the session was broken before receiving any "Resync Partial Message" or "Resync Finished Message". + +If one of these 5 seconds timeout expire, the process will try to request resync from a remote connected peer (see b). The process will wait until 5seconds +if no available remote peers are found. + +If the timeout expire, the process will consider itself ass fully updated + +b) Resync from remote peers + +The process will randomly choose a remote connected peer and ask for a full resync using a Resync Request Message. The process will wait until 5seconds +if no available remote peers are found. + +The chosen remote peer will push its all known data ending with a Resync Finished Message or a Resync Partial Message (if it it does not consider itself as full updated). + +If it receives a Resync Finished Message it will consider itself as fully updated and stops to ask for resync. + +If it receives a Resync Partial Message, the current peer will be flagged to anymore be requested and any other connected peer will be randomly chosen for a resync request (5s). + +If the session is broken before receiving any of these messages any other connected peer will be randomly chosen for a resync request (5s). + +If the timeout expire, the process will consider itself as fully updated + + |