| 1 |
0.7.0 |
| 2 |
===== |
| 3 |
|
| 4 |
Features |
| 5 |
-------- |
| 6 |
- Secondary indexes (indexes on column values) are now supported |
| 7 |
- Row size limit increased from 2GB to 2 billion columns. rows |
| 8 |
are no longer read into memory during compaction. |
| 9 |
- Keyspace and ColumnFamily definitions may be added and modified live |
| 10 |
- Streaming data for repair or node movement no longer requires |
| 11 |
anticompaction step first |
| 12 |
- NetworkTopologyStrategy (formerly DatacenterShardStrategy) is ready for |
| 13 |
use, enabling ConsistencyLevel.DCQUORUM and DCQUORUMSYNC. See comments |
| 14 |
in `cassandra.yaml.` |
| 15 |
- Optional per-Column time-to-live field allows expiring data without |
| 16 |
have to issue explicit remove commands |
| 17 |
- `truncate` thrift method allows clearing an entire ColumnFamily at once |
| 18 |
- Hadoop OutputFormat and Streaming [non-jvm map/reduce via stdin/out] |
| 19 |
support |
| 20 |
- Up to 8x faster reads from row cache |
| 21 |
- A new ByteOrderedPartitioner supports bytes keys with arbitrary content, |
| 22 |
and orders keys by their byte value. This should be used in new |
| 23 |
deployments instead of OrderPreservingPartitioner. |
| 24 |
- Optional round-robin scheduling between keyspaces for multitenant |
| 25 |
clusters |
| 26 |
- Dynamic endpoint snitch mitigates the impact of impaired nodes |
| 27 |
- New `IntegerType`, faster than LongType and allows integers of |
| 28 |
both less and more bits than Long's 64 |
| 29 |
- A revamped authentication system that decouples authorization and |
| 30 |
allows finer-grained control of resources. |
| 31 |
|
| 32 |
Upgrading |
| 33 |
--------- |
| 34 |
The Thrift API has changed in incompatible ways; see below, and refer |
| 35 |
to http://wiki.apache.org/cassandra/ClientOptions for a list of |
| 36 |
higher-level clients that have been updated to support the 0.7 API. |
| 37 |
|
| 38 |
The Cassandra inter-node protocol is incompatible with 0.6.x |
| 39 |
releases (and with 0.7 beta1), meaning you will have to bring your |
| 40 |
cluster down prior to upgrading: you cannot mix 0.6 and 0.7 nodes. |
| 41 |
|
| 42 |
The hints schema was changed from 0.6 to 0.7. Cassandra automatically |
| 43 |
snapshots and then truncates the hints column family as part of |
| 44 |
starting up 0.7 for the first time. |
| 45 |
|
| 46 |
Keyspace and ColumnFamily definitions are stored in the system |
| 47 |
keyspace, rather than the configuration file. |
| 48 |
|
| 49 |
The process to upgrade is: |
| 50 |
1) run "nodetool drain" on _each_ 0.6 node. When drain finishes (log |
| 51 |
message "Node is drained" appears), stop the process. |
| 52 |
2) Convert your storage-conf.xml to the new cassandra.yaml using |
| 53 |
"bin/config-converter". |
| 54 |
3) Rename any of your keyspace or column family names that do not adhere |
| 55 |
to the '^\w+' regex convention. |
| 56 |
4) Start up your cluster with the 0.7 version. |
| 57 |
5) Initialize your Keyspace and ColumnFamily definitions using |
| 58 |
"bin/schematool <host> <jmxport> import". _You only need to do |
| 59 |
this to one node_. |
| 60 |
|
| 61 |
Thrift API |
| 62 |
---------- |
| 63 |
- The Cassandra server now defaults to framed mode, rather than |
| 64 |
unframed. Unframed is obsolete and will be removed in the future. |
| 65 |
- The Cassandra Thrift interface file has been updated for Thrift 0.5. |
| 66 |
If you are compiling your own client code from the interface, you |
| 67 |
will need to upgrade the Thrift compiler to match. |
| 68 |
- Row keys are now bytes: keys stored by versions prior to 0.7.0 will be |
| 69 |
returned as UTF-8 encoded bytes. OrderPreservingPartitioner and |
| 70 |
CollatingOrderPreservingPartitioner continue to expect that keys contain |
| 71 |
UTF-8 encoded strings, but RandomPartitioner now works on any key data. |
| 72 |
- keyspace parameters have been replaced with the per-connection |
| 73 |
set_keyspace method. |
| 74 |
- The return type for login() is now AccessLevel. |
| 75 |
- The get_string_property() method has been removed. |
| 76 |
- The get_string_list_property() method has been removed. |
| 77 |
|
| 78 |
Configuraton |
| 79 |
------------ |
| 80 |
- Configuration file renamed to cassandra.yaml and log4j.properties to |
| 81 |
log4j-server.properties |
| 82 |
- PropertyFileSnitch configuration file renamed to |
| 83 |
cassandra-topology.properties |
| 84 |
- The ThriftAddress and ThriftPort directives have been renamed to |
| 85 |
RPCAddress and RPCPort respectively. |
| 86 |
- EndPointSnitch was renamed to RackInferringSnitch. A new SimpleSnitch |
| 87 |
has been added. |
| 88 |
- RackUnawareStrategy and RackAwareStrategy have been renamed to |
| 89 |
SimpleStrategy and OldNetworkTopologyStrategy, respectively. |
| 90 |
- RowWarningThresholdInMB replaced with in_memory_compaction_limit_in_mb |
| 91 |
- GCGraceSeconds is now per-ColumnFamily instead of global |
| 92 |
- Keyspace and column family names that do not confirm to a '^\w+' regex |
| 93 |
are considered illegal. |
| 94 |
- Keyspace and column family definitions will need to be loaded via |
| 95 |
"bin/schematool <host> <jmxport> import". _You only need to do this to |
| 96 |
one node_. |
| 97 |
- In addition to an authenticator, an authority must be configured as |
| 98 |
well. Users of SimpleAuthenticator should use SimpleAuthority for this |
| 99 |
value (the default is AllowAllAuthority, which corresponds with |
| 100 |
AllowAllAuthenticator). |
| 101 |
- The format of access.properties has changed, see the sample configuration |
| 102 |
conf/access.properties for documentation on the new format. |
| 103 |
|
| 104 |
|
| 105 |
JMX |
| 106 |
--- |
| 107 |
- StreamingService moved from o.a.c.streaming to o.a.c.service |
| 108 |
- GMFD renamed to GOSSIP_STAGE |
| 109 |
- {Min,Mean,Max}RowCompactedSize renamed to {Min,Mean,Max}RowSize |
| 110 |
since it no longer has to wait til compaction to be computed |
| 111 |
|
| 112 |
Other |
| 113 |
----- |
| 114 |
- If extending AbstractType, make sure you follow the singleton pattern |
| 115 |
followed by Cassandra core AbstractType classes: provide a public |
| 116 |
static final variable called 'instance'. |
| 117 |
|
| 118 |
|
| 119 |
0.6.6 |
| 120 |
===== |
| 121 |
|
| 122 |
Upgrading |
| 123 |
--------- |
| 124 |
- As part of the cache-saving feature, a third directory |
| 125 |
(along with data and commitlog) has been added to the config |
| 126 |
file. You will need to set and create this directory |
| 127 |
when restarting your node into 0.6.6. |
| 128 |
|
| 129 |
|
| 130 |
0.6.1 |
| 131 |
===== |
| 132 |
|
| 133 |
Upgrading |
| 134 |
--------- |
| 135 |
- We try to keep minor versions 100% compatible (data format, |
| 136 |
commitlog format, network format) within the major series, but |
| 137 |
we introduced a network-level incompatibility in this 0.6.1. |
| 138 |
Thus, if you are upgrading from 0.6.0 to any higher version |
| 139 |
(0.6.1, 0.6.2, etc.) then you will need to restart your entire |
| 140 |
cluster with the new version, instead of being able to do a |
| 141 |
rolling restart. |
| 142 |
|
| 143 |
|
| 144 |
0.6.0 |
| 145 |
===== |
| 146 |
|
| 147 |
Features |
| 148 |
-------- |
| 149 |
- row caching: configure with the RowsCached attribute in |
| 150 |
ColumnFamily definition |
| 151 |
- Hadoop map/reduce support: see contrib/word_count for an example |
| 152 |
- experimental authentication support, described under |
| 153 |
Authenticator in storage.conf |
| 154 |
|
| 155 |
Configuraton |
| 156 |
------------ |
| 157 |
- MemtableSizeInMB has been replaced by MemtableThroughputInMB which |
| 158 |
triggers a memtable flush when the specified amount of data has |
| 159 |
been written, including overwrites. |
| 160 |
- MemtableObjectCountInMillions has been replaced by the |
| 161 |
MemtableOperationsInMillions directive which causes a memtable flush |
| 162 |
to occur after the specified number of operations. |
| 163 |
- Like MemtableSizeInMB, BinaryMemtableSizeInMB has been replaced by |
| 164 |
BinaryMemtableThroughputInMB. |
| 165 |
- Replication factor is now per-keyspace, rather than global. |
| 166 |
- KeysCachedFraction is deprecated in favor of KeysCached |
| 167 |
- RowWarningThresholdInMB added, to warn before very large rows |
| 168 |
get big enough to threaten node stability |
| 169 |
|
| 170 |
Thrift API |
| 171 |
---------- |
| 172 |
- removed deprecated get_key_range method |
| 173 |
- added batch_mutate meethod |
| 174 |
- deprecated multiget and batch_insert methods in favor of |
| 175 |
multiget_slice and batch_mutate, respectively |
| 176 |
- added ConsistencyLevel.ANY, for when you want write |
| 177 |
availability even when it may not be readable immediately. |
| 178 |
Unlike CL.ZERO, though, it will throw an exception if |
| 179 |
it cannot be written *somewhere*. |
| 180 |
|
| 181 |
JMX metrics |
| 182 |
----------- |
| 183 |
- read and write statistics are reported as lifetime totals, |
| 184 |
instead of averages over the last minute. average-since-last |
| 185 |
requested are also available for convenience. |
| 186 |
- cache hit rate statistics are now available from JMX under |
| 187 |
org.apache.cassandra.db.Caches |
| 188 |
- compaction JMX metrics are moved to |
| 189 |
org.apache.cassandra.db.CompactionManager. PendingTasks is now |
| 190 |
a much better estimate of compactions remaining, and the |
| 191 |
progress of the current compaction has been added. |
| 192 |
- commitlog JMX metrics are moved to org.apache.cassandra.db.Commitlog |
| 193 |
- progress of data streaming during bootstrap, loadbalance, or other |
| 194 |
data migration, is available under |
| 195 |
org.apache.cassandra.streaming.StreamingService. |
| 196 |
See http://wiki.apache.org/cassandra/Streaming for details. |
| 197 |
|
| 198 |
Installation/Upgrade |
| 199 |
-------------------- |
| 200 |
- 0.6 network traffic is not compatible with earlier versions. You |
| 201 |
will need to shut down all your nodes at once, upgrade, then restart. |
| 202 |
|
| 203 |
|
| 204 |
|
| 205 |
0.5.0 |
| 206 |
===== |
| 207 |
|
| 208 |
0. The commitlog format has changed (but sstable format has not). |
| 209 |
When upgrading from 0.4, empty the commitlog either by running |
| 210 |
bin/nodeprobe flush on each machine and waiting for the flush to finish, |
| 211 |
or simply remove the commitlog directory if you only have test data. |
| 212 |
(If more writes come in after the flush command, starting 0.5 will error |
| 213 |
out; if that happens, just go back to 0.4 and flush again.) |
| 214 |
The format changed twice: from 0.4 to beta1, and from beta2 to RC1. |
| 215 |
|
| 216 |
.5 The gossip protocol has changed, meaning 0.5 nodes cannot coexist |
| 217 |
in a cluster of 0.4 nodes or vice versa; you must upgrade your |
| 218 |
whole cluster at the same time. |
| 219 |
|
| 220 |
1. Bootstrap, move, load balancing, and active repair have been added. |
| 221 |
See http://wiki.apache.org/cassandra/Operations. When upgrading |
| 222 |
from 0.4, leave autobootstrap set to false for the first restart |
| 223 |
of your old nodes. |
| 224 |
|
| 225 |
2. Performance improvements across the board, especially on the write |
| 226 |
path (over 100% improvement in stress.py throughput). |
| 227 |
|
| 228 |
3. Configuration: |
| 229 |
- Added "comment" field to ColumnFamily definition. |
| 230 |
- Added MemtableFlushAfterMinutes, a global replacement for the |
| 231 |
old per-CF FlushPeriodInMinutes setting |
| 232 |
- Key cache settings |
| 233 |
|
| 234 |
4. Thrift: |
| 235 |
- Added get_range_slice, deprecating get_key_range |
| 236 |
|
| 237 |
|
| 238 |
|
| 239 |
0.4.2 |
| 240 |
===== |
| 241 |
|
| 242 |
1. Improve default garbage collector options significantly -- |
| 243 |
throughput will be 30% higher or more. |
| 244 |
|
| 245 |
|
| 246 |
|
| 247 |
0.4.1 |
| 248 |
===== |
| 249 |
|
| 250 |
1. SnapshotBeforeCompaction configuration option allows snapshotting |
| 251 |
before each compaction, which allows rolling back to any version |
| 252 |
of the data. |
| 253 |
|
| 254 |
|
| 255 |
|
| 256 |
0.4.0 |
| 257 |
===== |
| 258 |
|
| 259 |
1. On-disk data format has changed to allow billions of keys/rows per |
| 260 |
node instead of only millions. The new format is incompatible with 0.3; |
| 261 |
see 0.3 notes below for how to import data from a 0.3 install. |
| 262 |
|
| 263 |
2. Cassandra now supports multiple keyspaces. Typically you will have |
| 264 |
one keyspace per application, allowing applications to be able to |
| 265 |
create and modify ColumnFamilies at will without worrying about |
| 266 |
collisions with others in the same cluster. |
| 267 |
|
| 268 |
3. Many Thrift API changes and documentation. See |
| 269 |
http://wiki.apache.org/cassandra/API |
| 270 |
|
| 271 |
4. Removed the web interface in favor of JMX and bin/nodeprobe, which |
| 272 |
has significantly enhanced functionality. |
| 273 |
|
| 274 |
5. Renamed configuration "<Table>" to "<Keyspace>". |
| 275 |
|
| 276 |
6. Added commitlog fsync; see "<CommitLogSync>" in configuration. |
| 277 |
|
| 278 |
|
| 279 |
|
| 280 |
0.3.0 |
| 281 |
===== |
| 282 |
|
| 283 |
1. With enough and large enough keys in a ColumnFamily, Cassandra will |
| 284 |
run out of memory trying to perform compactions (data file merges). |
| 285 |
The size of what is stored in memory is (S + 16) * (N + M) where S |
| 286 |
is the size of the key (usually 2 bytes per character), N is the |
| 287 |
number of keys and M, is the map overhead (which can be guestimated |
| 288 |
at around 32 bytes per key). |
| 289 |
So, if you have 10-character keys and 1GB of headroom in your heap |
| 290 |
space for compaction, you can expect to store about 17M keys |
| 291 |
before running into problems. |
| 292 |
See https://issues.apache.org/jira/browse/CASSANDRA-208 |
| 293 |
|
| 294 |
2. Because fixing #1 requires a data file format change, 0.4 will not |
| 295 |
be binary-compatible with 0.3 data files. A client-side upgrade |
| 296 |
can be done relatively easily with the following algorithm: |
| 297 |
for key in old_client.get_key_range(everything): |
| 298 |
columns = old_client.get_slice or get_slice_super(key, all columns) |
| 299 |
new_client.batch_insert or batch_insert_super(key, columns) |
| 300 |
The inner loop can be trivially parallelized for speed. |
| 301 |
|
| 302 |
3. Commitlog does not fsync before reporting a write successful. |
| 303 |
Using blocking writes mitigates this to some degree, since all |
| 304 |
nodes that were part of the write quorum would have to fail |
| 305 |
before sync for data to be lost. |
| 306 |
See https://issues.apache.org/jira/browse/CASSANDRA-182 |
| 307 |
|
| 308 |
Additionally, row size (that is, all the data associated with a single |
| 309 |
key in a given ColumnFamily) is limited by available memory, because |
| 310 |
compaction deserializes each row before merging. |
| 311 |
|
| 312 |
See https://issues.apache.org/jira/browse/CASSANDRA-16 |
| 313 |
|
Parent Directory
| 
Revision Log