docs/COLO-FT.txt

   1 COarse-grained LOck-stepping Virtual Machines for Non-stop Service
   2 ----------------------------------------
   3 Copyright (c) 2016 Intel Corporation
   4 Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
   5 Copyright (c) 2016 Fujitsu, Corp.
   6
   7 This work is licensed under the terms of the GNU GPL, version 2 or later.
   8 See the COPYING file in the top-level directory.
   9
  10 This document gives an overview of COLO's design and how to use it.
  11
  12 == Background ==
  13 Virtual machine (VM) replication is a well known technique for providing
  14 application-agnostic software-implemented hardware fault tolerance,
  15 also known as "non-stop service".
  16
  17 COLO (COarse-grained LOck-stepping) is a high availability solution.
  18 Both primary VM (PVM) and secondary VM (SVM) run in parallel. They receive the
  19 same request from client, and generate response in parallel too.
  20 If the response packets from PVM and SVM are identical, they are released
  21 immediately. Otherwise, a VM checkpoint (on demand) is conducted.
  22
  23 == Architecture ==
  24
  25 The architecture of COLO is shown in the diagram below.
  26 It consists of a pair of networked physical nodes:
  27 The primary node running the PVM, and the secondary node running the SVM
  28 to maintain a valid replica of the PVM.
  29 PVM and SVM execute in parallel and generate output of response packets for
  30 client requests according to the application semantics.
  31
  32 The incoming packets from the client or external network are received by the
  33 primary node, and then forwarded to the secondary node, so that both the PVM
  34 and the SVM are stimulated with the same requests.
  35
  36 COLO receives the outbound packets from both the PVM and SVM and compares them
  37 before allowing the output to be sent to clients.
  38
  39 The SVM is qualified as a valid replica of the PVM, as long as it generates
  40 identical responses to all client requests. Once the differences in the outputs
  41 are detected between the PVM and SVM, COLO withholds transmission of the
  42 outbound packets until it has successfully synchronized the PVM state to the SVM.
  43
  44   Primary Node                                                            Secondary Node
  45 +------------+  +-----------------------+       +------------------------+  +------------+
  46 |            |  |       HeartBeat       +<----->+       HeartBeat        |  |            |
  47 | Primary VM |  +-----------+-----------+       +-----------+------------+  |Secondary VM|
  48 |            |              |                               |               |            |
  49 |            |  +-----------|-----------+       +-----------|------------+  |            |
  50 |            |  |QEMU   +---v----+      |       |QEMU  +----v---+        |  |            |
  51 |            |  |       |Failover|      |       |      |Failover|        |  |            |
  52 |            |  |       +--------+      |       |      +--------+        |  |            |
  53 |            |  |   +---------------+   |       |   +---------------+    |  |            |
  54 |            |  |   | VM Checkpoint +-------------->+ VM Checkpoint |    |  |            |
  55 |            |  |   +---------------+   |       |   +---------------+    |  |            |
  56 |Requests<--------------------------\ /-----------------\ /--------------------->Requests|
  57 |            |  |                   ^ ^ |       |       | |              |  |            |
  58 |Responses+---------------------\ /-|-|------------\ /-------------------------+Responses|
  59 |            |  |               | | | | |       |  | |  | |              |  |            |
  60 |            |  | +-----------+ | | | | |       |  | |  | | +----------+ |  |            |
  61 |            |  | | COLO disk | | | | | |       |  | |  | | | COLO disk| |  |            |
  62 |            |  | |   Manager +---------------------------->| Manager  | |  |            |
  63 |            |  | ++----------+ v v | | |       |  | v  v | +---------++ |  |            |
  64 |            |  |  |+-----------+-+-+-++|       | ++-+--+-+---------+ |  |  |            |
  65 |            |  |  ||   COLO Proxy     ||       | |   COLO Proxy    | |  |  |            |
  66 |            |  |  || (compare packet  ||       | |(adjust sequence | |  |  |            |
  67 |            |  |  ||and mirror packet)||       | |    and ACK)     | |  |  |            |
  68 |            |  |  |+------------+---+-+|       | +-----------------+ |  |  |            |
  69 +------------+  +-----------------------+       +------------------------+  +------------+
  70 +------------+     |             |   |                                |     +------------+
  71 | VM Monitor |     |             |   |                                |     | VM Monitor |
  72 +------------+     |             |   |                                |     +------------+
  73 +---------------------------------------+       +----------------------------------------+
  74 |   Kernel         |             |   |  |       |   Kernel            |                  |
  75 +---------------------------------------+       +----------------------------------------+
  76                    |             |   |                                |
  77     +--------------v+  +---------v---+--+       +------------------+ +v-------------+
  78     |   Storage     |  |External Network|       | External Network | |   Storage    |
  79     +---------------+  +----------------+       +------------------+ +--------------+
  80
  81
  82 == Components introduction ==
  83
  84 You can see there are several components in COLO's diagram of architecture.
  85 Their functions are described below.
  86
  87 HeartBeat:
  88 Runs on both the primary and secondary nodes, to periodically check platform
  89 availability. When the primary node suffers a hardware fail-stop failure,
  90 the heartbeat stops responding, the secondary node will trigger a failover
  91 as soon as it determines the absence.
  92
  93 COLO disk Manager:
  94 When primary VM writes data into image, the colo disk manager captures this data
  95 and sends it to secondary VM's which makes sure the context of secondary VM's
  96 image is consistent with the context of primary VM 's image.
  97 For more details, please refer to docs/block-replication.txt.
  98
  99 Checkpoint/Failover Controller:
 100 Modifications of save/restore flow to realize continuous migration,
 101 to make sure the state of VM in Secondary side is always consistent with VM in
 102 Primary side.
 103
 104 COLO Proxy:
 105 Delivers packets to Primary and Secondary, and then compare the responses from
 106 both side. Then decide whether to start a checkpoint according to some rules.
 107 Please refer to docs/colo-proxy.txt for more information.
 108
 109 Note:
 110 HeartBeat has not been implemented yet, so you need to trigger failover process
 111 by using 'x-colo-lost-heartbeat' command.
 112
 113 == COLO operation status ==
 114
 115 +-----------------+
 116 |                 |
 117 |    Start COLO   |
 118 |                 |
 119 +--------+--------+
 120          |
 121          |  Main qmp command:
 122          |  migrate-set-capabilities with x-colo
 123          |  migrate
 124          |
 125          v
 126 +--------+--------+
 127 |                 |
 128 |  COLO running   |
 129 |                 |
 130 +--------+--------+
 131          |
 132          |  Main qmp command:
 133          |  x-colo-lost-heartbeat
 134          |  or
 135          |  some error happened
 136          v
 137 +--------+--------+
 138 |                 |  send qmp event:
 139 |  COLO failover  |  COLO_EXIT
 140 |                 |
 141 +-----------------+
 142
 143 COLO use the qmp command to switch and report operation status.
 144 The diagram just shows the main qmp command, you can get the detail
 145 in test procedure.
 146
 147 == Test procedure ==
 148 Note: Here we are running both instances on the same host for testing,
 149 change the IP Addresses if you want to run it on two hosts. Initially
 150 127.0.0.1 is the Primary Host and 127.0.0.2 is the Secondary Host.
 151
 152 == Startup qemu ==
 153 1. Primary:
 154 Note: Initially, $imagefolder/primary.qcow2 needs to be copied to all hosts.
 155 You don't need to change any IP's here, because 0.0.0.0 listens on any
 156 interface. The chardev's with 127.0.0.1 IP's loopback to the local qemu
 157 instance.
 158
 159 # imagefolder="/mnt/vms/colo-test-primary"
 160
 161 # qemu-system-x86_64 -enable-kvm -cpu qemu64,kvmclock=on -m 512 -smp 1 -qmp stdio \
 162    -device piix3-usb-uhci -device usb-tablet -name primary \
 163    -netdev tap,id=hn0,vhost=off,helper=/usr/lib/qemu/qemu-bridge-helper \
 164    -device rtl8139,id=e0,netdev=hn0 \
 165    -chardev socket,id=mirror0,host=0.0.0.0,port=9003,server=on,wait=off \
 166    -chardev socket,id=compare1,host=0.0.0.0,port=9004,server=on,wait=on \
 167    -chardev socket,id=compare0,host=127.0.0.1,port=9001,server=on,wait=off \
 168    -chardev socket,id=compare0-0,host=127.0.0.1,port=9001 \
 169    -chardev socket,id=compare_out,host=127.0.0.1,port=9005,server=on,wait=off \
 170    -chardev socket,id=compare_out0,host=127.0.0.1,port=9005 \
 171    -object filter-mirror,id=m0,netdev=hn0,queue=tx,outdev=mirror0 \
 172    -object filter-redirector,netdev=hn0,id=redire0,queue=rx,indev=compare_out \
 173    -object filter-redirector,netdev=hn0,id=redire1,queue=rx,outdev=compare0 \
 174    -object iothread,id=iothread1 \
 175    -object colo-compare,id=comp0,primary_in=compare0-0,secondary_in=compare1,\
 176 outdev=compare_out0,iothread=iothread1 \
 177    -drive if=ide,id=colo-disk0,driver=quorum,read-pattern=fifo,vote-threshold=1,\
 178 children.0.file.filename=$imagefolder/primary.qcow2,children.0.driver=qcow2 -S
 179
 180 2. Secondary:
 181 Note: Active and hidden images need to be created only once and the
 182 size should be the same as primary.qcow2. Again, you don't need to change
 183 any IP's here, except for the $primary_ip variable.
 184
 185 # imagefolder="/mnt/vms/colo-test-secondary"
 186 # primary_ip=127.0.0.1
 187
 188 # qemu-img create -f qcow2 $imagefolder/secondary-active.qcow2 10G
 189
 190 # qemu-img create -f qcow2 $imagefolder/secondary-hidden.qcow2 10G
 191
 192 # qemu-system-x86_64 -enable-kvm -cpu qemu64,kvmclock=on -m 512 -smp 1 -qmp stdio \
 193    -device piix3-usb-uhci -device usb-tablet -name secondary \
 194    -netdev tap,id=hn0,vhost=off,helper=/usr/lib/qemu/qemu-bridge-helper \
 195    -device rtl8139,id=e0,netdev=hn0 \
 196    -chardev socket,id=red0,host=$primary_ip,port=9003,reconnect=1 \
 197    -chardev socket,id=red1,host=$primary_ip,port=9004,reconnect=1 \
 198    -object filter-redirector,id=f1,netdev=hn0,queue=tx,indev=red0 \
 199    -object filter-redirector,id=f2,netdev=hn0,queue=rx,outdev=red1 \
 200    -object filter-rewriter,id=rew0,netdev=hn0,queue=all \
 201    -drive if=none,id=parent0,file.filename=$imagefolder/primary.qcow2,driver=qcow2 \
 202    -drive if=none,id=childs0,driver=replication,mode=secondary,file.driver=qcow2,\
 203 top-id=colo-disk0,file.file.filename=$imagefolder/secondary-active.qcow2,\
 204 file.backing.driver=qcow2,file.backing.file.filename=$imagefolder/secondary-hidden.qcow2,\
 205 file.backing.backing=parent0 \
 206    -drive if=ide,id=colo-disk0,driver=quorum,read-pattern=fifo,vote-threshold=1,\
 207 children.0=childs0 \
 208    -incoming tcp:0.0.0.0:9998
 209
 210
 211 3. On Secondary VM's QEMU monitor, issue command
 212 {"execute":"qmp_capabilities"}
 213 {"execute": "migrate-set-capabilities", "arguments": {"capabilities": [ {"capability": "x-colo", "state": true } ] } }
 214 {"execute": "nbd-server-start", "arguments": {"addr": {"type": "inet", "data": {"host": "0.0.0.0", "port": "9999"} } } }
 215 {"execute": "nbd-server-add", "arguments": {"device": "parent0", "writable": true } }
 216
 217 Note:
 218   a. The qmp command nbd-server-start and nbd-server-add must be run
 219      before running the qmp command migrate on primary QEMU
 220   b. Active disk, hidden disk and nbd target's length should be the
 221      same.
 222   c. It is better to put active disk and hidden disk in ramdisk. They
 223      will be merged into the parent disk on failover.
 224
 225 4. On Primary VM's QEMU monitor, issue command:
 226 {"execute":"qmp_capabilities"}
 227 {"execute": "human-monitor-command", "arguments": {"command-line": "drive_add -n buddy driver=replication,mode=primary,file.driver=nbd,file.host=127.0.0.2,file.port=9999,file.export=parent0,node-name=replication0"}}
 228 {"execute": "x-blockdev-change", "arguments":{"parent": "colo-disk0", "node": "replication0" } }
 229 {"execute": "migrate-set-capabilities", "arguments": {"capabilities": [ {"capability": "x-colo", "state": true } ] } }
 230 {"execute": "migrate", "arguments": {"uri": "tcp:127.0.0.2:9998" } }
 231
 232   Note:
 233   a. There should be only one NBD Client for each primary disk.
 234   b. The qmp command line must be run after running qmp command line in
 235      secondary qemu.
 236
 237 5. After the above steps, you will see, whenever you make changes to PVM, SVM will be synced.
 238 You can issue command '{ "execute": "migrate-set-parameters" , "arguments":{ "x-checkpoint-delay": 2000 } }'
 239 to change the idle checkpoint period time
 240
 241 6. Failover test
 242 You can kill one of the VMs and Failover on the surviving VM:
 243
 244 If you killed the Secondary, then follow "Primary Failover". After that,
 245 if you want to resume the replication, follow "Primary resume replication"
 246
 247 If you killed the Primary, then follow "Secondary Failover". After that,
 248 if you want to resume the replication, follow "Secondary resume replication"
 249
 250 == Primary Failover ==
 251 The Secondary died, resume on the Primary
 252
 253 {"execute": "x-blockdev-change", "arguments":{ "parent": "colo-disk0", "child": "children.1"} }
 254 {"execute": "human-monitor-command", "arguments":{ "command-line": "drive_del replication0" } }
 255 {"execute": "object-del", "arguments":{ "id": "comp0" } }
 256 {"execute": "object-del", "arguments":{ "id": "iothread1" } }
 257 {"execute": "object-del", "arguments":{ "id": "m0" } }
 258 {"execute": "object-del", "arguments":{ "id": "redire0" } }
 259 {"execute": "object-del", "arguments":{ "id": "redire1" } }
 260 {"execute": "x-colo-lost-heartbeat" }
 261
 262 == Secondary Failover ==
 263 The Primary died, resume on the Secondary and prepare to become the new Primary
 264
 265 {"execute": "nbd-server-stop"}
 266 {"execute": "x-colo-lost-heartbeat"}
 267
 268 {"execute": "object-del", "arguments":{ "id": "f2" } }
 269 {"execute": "object-del", "arguments":{ "id": "f1" } }
 270 {"execute": "chardev-remove", "arguments":{ "id": "red1" } }
 271 {"execute": "chardev-remove", "arguments":{ "id": "red0" } }
 272
 273 {"execute": "chardev-add", "arguments":{ "id": "mirror0", "backend": {"type": "socket", "data": {"addr": { "type": "inet", "data": { "host": "0.0.0.0", "port": "9003" } }, "server": true } } } }
 274 {"execute": "chardev-add", "arguments":{ "id": "compare1", "backend": {"type": "socket", "data": {"addr": { "type": "inet", "data": { "host": "0.0.0.0", "port": "9004" } }, "server": true } } } }
 275 {"execute": "chardev-add", "arguments":{ "id": "compare0", "backend": {"type": "socket", "data": {"addr": { "type": "inet", "data": { "host": "127.0.0.1", "port": "9001" } }, "server": true } } } }
 276 {"execute": "chardev-add", "arguments":{ "id": "compare0-0", "backend": {"type": "socket", "data": {"addr": { "type": "inet", "data": { "host": "127.0.0.1", "port": "9001" } }, "server": false } } } }
 277 {"execute": "chardev-add", "arguments":{ "id": "compare_out", "backend": {"type": "socket", "data": {"addr": { "type": "inet", "data": { "host": "127.0.0.1", "port": "9005" } }, "server": true } } } }
 278 {"execute": "chardev-add", "arguments":{ "id": "compare_out0", "backend": {"type": "socket", "data": {"addr": { "type": "inet", "data": { "host": "127.0.0.1", "port": "9005" } }, "server": false } } } }
 279
 280 == Primary resume replication ==
 281 Resume replication after new Secondary is up.
 282
 283 Start the new Secondary (Steps 2 and 3 above), then on the Primary:
 284 {"execute": "drive-mirror", "arguments":{ "device": "colo-disk0", "job-id": "resync", "target": "nbd://127.0.0.2:9999/parent0", "mode": "existing", "format": "raw", "sync": "full"} }
 285
 286 Wait until disk is synced, then:
 287 {"execute": "stop"}
 288 {"execute": "block-job-cancel", "arguments":{ "device": "resync"} }
 289
 290 {"execute": "human-monitor-command", "arguments":{ "command-line": "drive_add -n buddy driver=replication,mode=primary,file.driver=nbd,file.host=127.0.0.2,file.port=9999,file.export=parent0,node-name=replication0"}}
 291 {"execute": "x-blockdev-change", "arguments":{ "parent": "colo-disk0", "node": "replication0" } }
 292
 293 {"execute": "object-add", "arguments":{ "qom-type": "filter-mirror", "id": "m0", "netdev": "hn0", "queue": "tx", "outdev": "mirror0" } }
 294 {"execute": "object-add", "arguments":{ "qom-type": "filter-redirector", "id": "redire0", "netdev": "hn0", "queue": "rx", "indev": "compare_out" } }
 295 {"execute": "object-add", "arguments":{ "qom-type": "filter-redirector", "id": "redire1", "netdev": "hn0", "queue": "rx", "outdev": "compare0" } }
 296 {"execute": "object-add", "arguments":{ "qom-type": "iothread", "id": "iothread1" } }
 297 {"execute": "object-add", "arguments":{ "qom-type": "colo-compare", "id": "comp0", "primary_in": "compare0-0", "secondary_in": "compare1", "outdev": "compare_out0", "iothread": "iothread1" } }
 298
 299 {"execute": "migrate-set-capabilities", "arguments":{ "capabilities": [ {"capability": "x-colo", "state": true } ] } }
 300 {"execute": "migrate", "arguments":{ "uri": "tcp:127.0.0.2:9998" } }
 301
 302 Note:
 303 If this Primary previously was a Secondary, then we need to insert the
 304 filters before the filter-rewriter by using the
 305 ""insert": "before", "position": "id=rew0"" Options. See below.
 306
 307 == Secondary resume replication ==
 308 Become Primary and resume replication after new Secondary is up. Note
 309 that now 127.0.0.1 is the Secondary and 127.0.0.2 is the Primary.
 310
 311 Start the new Secondary (Steps 2 and 3 above, but with primary_ip=127.0.0.2),
 312 then on the old Secondary:
 313 {"execute": "drive-mirror", "arguments":{ "device": "colo-disk0", "job-id": "resync", "target": "nbd://127.0.0.1:9999/parent0", "mode": "existing", "format": "raw", "sync": "full"} }
 314
 315 Wait until disk is synced, then:
 316 {"execute": "stop"}
 317 {"execute": "block-job-cancel", "arguments":{ "device": "resync" } }
 318
 319 {"execute": "human-monitor-command", "arguments":{ "command-line": "drive_add -n buddy driver=replication,mode=primary,file.driver=nbd,file.host=127.0.0.1,file.port=9999,file.export=parent0,node-name=replication0"}}
 320 {"execute": "x-blockdev-change", "arguments":{ "parent": "colo-disk0", "node": "replication0" } }
 321
 322 {"execute": "object-add", "arguments":{ "qom-type": "filter-mirror", "id": "m0", "insert": "before", "position": "id=rew0", "netdev": "hn0", "queue": "tx", "outdev": "mirror0" } }
 323 {"execute": "object-add", "arguments":{ "qom-type": "filter-redirector", "id": "redire0", "insert": "before", "position": "id=rew0", "netdev": "hn0", "queue": "rx", "indev": "compare_out" } }
 324 {"execute": "object-add", "arguments":{ "qom-type": "filter-redirector", "id": "redire1", "insert": "before", "position": "id=rew0", "netdev": "hn0", "queue": "rx", "outdev": "compare0" } }
 325 {"execute": "object-add", "arguments":{ "qom-type": "iothread", "id": "iothread1" } }
 326 {"execute": "object-add", "arguments":{ "qom-type": "colo-compare", "id": "comp0", "primary_in": "compare0-0", "secondary_in": "compare1", "outdev": "compare_out0", "iothread": "iothread1" } }
 327
 328 {"execute": "migrate-set-capabilities", "arguments":{ "capabilities": [ {"capability": "x-colo", "state": true } ] } }
 329 {"execute": "migrate", "arguments":{ "uri": "tcp:127.0.0.1:9998" } }
 330
 331 == TODO ==
 332 1. Support shared storage.
 333 2. Develop the heartbeat part.
 334 3. Reduce checkpoint VM’s downtime while doing checkpoint.