Overview
RGW 中三个基本概念：user， bucket， object。通过分析RGW data layout，可以清楚对象存储的三个基本概念是怎样在RGW 中实现的。
RGW 中数据分三种类型：
- data: 每个RGW object 会保存在一个或多个Rados object(s)
- metadata: user, bucket, bucket.instance
- bucket index: 单独作为一种metadata
RGW 的数据在RADOS 层以三种形式存在：rados对象（data 部分），rados 对象扩展属性xattr，rados 对象的omap中。
metadata
RGW metadata分为：
- user：保存user 信息
- bucket：维护一组bucket name 和bucket instance id 的映射
- bucket.instance：保存bucket instance 信息
- otp：N 版新增，otp (one-time password) 信息。根据虚拟或硬件Multi-Factor Authentication (MFA) 设备，基于已经时间同步的otp 算法生成一个密码。
metadata list，可以看到是上面列出的四种类型：
[root@stor14 build]# bin/radosgw-admin metadata list -c ceph.conf
[
"bucket",
"bucket.instance",
"otp",
"user"
]查看user 下都有什么 ：
[root@stor14 build]# bin/radosgw-admin metadata list user -c ceph.conf
[
"56789abcdef0123456789abcdef0123456789abcdef0123456789abcdef01234",
"bl_deliver",
"testx$9876543210abcdef0123456789abcdef0123456789abcdef0123456789abcdef",
"0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef",
"user1",
"test",
"testid",
"ms_sync"
]list bucket.instance 看下，一般一个bucket 对应一个bucket.instance
[root@stor14 build]# bin/radosgw-admin metadata list bucket.instance -c ceph.conf
[
"bltest1:e34456f0-d371-4384-9007-70c60563fb0b.4281.17",
"test1:e34456f0-d371-4384-9007-70c60563fb0b.4281.15"
]接着看下bltest1:e34456f0-d371-4384-9007-70c60563fb0b.4281.17 这个bucket.instance
[root@stor14 build]# bin/radosgw-admin metadata get bucket.instance:bltest1:e34456f0-d371-4384-9007-70c60563fb0b.4281.17 -c ceph.conf
{
"key": "bucket.instance:bltest1:e34456f0-d371-4384-9007-70c60563fb0b.4281.17",
"ver": {
"tag": "_UHAkf8NRdh_eXm7CqYsFNA8",
"ver": 1
},
"mtime": "2019-11-20 07:53:39.493618Z",
"data": {
"bucket_info": {
"bucket": {
"name": "bltest1",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4281.17",
"bucket_id": "e34456f0-d371-4384-9007-70c60563fb0b.4281.17",
"tenant": "",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
}
},
"creation_time": "2019-11-20 07:53:39.399284Z",
"owner": "user1",
"flags": 0,
"zonegroup": "513b96de-2450-4292-a86a-314abbe29766",
"placement_rule": "default-placement",
"has_instance_obj": "true",
"quota": {
"enabled": false,
"check_on_raw": false,
"max_size": -1,
"max_size_kb": 0,
"max_objects": -1
},
"num_shards": 0,
"bi_shard_hash_type": 0,
"requester_pays": "false",
"has_website": "false",
"swift_versioning": "false",
"swift_ver_location": "",
"index_type": 0,
"mdsearch_config": [],
"reshard_status": 0,
"new_bucket_instance_id": ""
},
"attrs": [
{
"key": "user.rgw.acl",
"val": "AgKBAAAAAwISAAAABQAAAHVzZXIxBQAAAHVzZXIxBANjAAAAAQEAAAAFAAAAdXNlcjEPAAAAAQAAAAUAAAB1c2VyMQUDNgAAAAICBAAAAAAAAAAFAAAAdXNlcjEAAAAAAAAAAAICBAAAAA8AAAAFAAAAdXNlcjEAAAAAAAAAAAAAAAAAAAAA"
}
]
}
}获取bltest1 这个存储桶相关信息：
[root@stor14 build]# bin/radosgw-admin metadata get bucket:bltest1 -c ceph.conf
{
"key": "bucket:bltest1",
"ver": {
"tag": "_Sia7A02mjQlrvVkymKa_y2G",
"ver": 1
},
"mtime": "2019-11-20 07:53:39.560897Z",
"data": {
"bucket": {
"name": "bltest1",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4281.17",
"bucket_id": "e34456f0-d371-4384-9007-70c60563fb0b.4281.17",
"tenant": "",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
}
},
"owner": "user1",
"creation_time": "2019-11-20 07:53:39.399284Z",
"linked": "true",
"has_bucket_info": "false"
}
}由于我们没有做MFA 相关配置，otp 目前为空：
[root@stor14 build]# bin/radosgw-admin metadata list otp -c ceph.conf
[]bucket index
bucket index 是一类特殊的metadata，通过bucket index 我们可以list 指定存储桶下的所有rgw 对象。
bucket index 对象保存在存储池
bucket index 维护了一个k-v map，
- k-v map本身保存在rados object(s) 关联的omap 中。在不启用shard 时，一个存储桶会对应一个rados object；若shard 后，一个存储桶可能会对应多个index rados objects。
- omap 的key 为各rgw object name，omap value 这些rgw object 的一些基本元数据，如list bucket 时展示的元数据。
- 每个omap 有一个header，在header 中保存一些bucket的统计信息（对象数，总大小等）?
list 一下bucket index 对象（rados object）：
[root@stor14 build]# bin/rados -c ceph.conf ls -p default.rgw.buckets.index
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4281.15
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4281.17可以看到两个index objects，此时default zone 下只有两个存储桶。
[root@stor14 build]# bin/radosgw-admin bucket list -c ceph.conf
[
"test1",
"bltest1"
]其中对应存储桶bltest1
[root@stor14 build]# bin/radosgw-admin bucket stats --bucket test1 -c ceph.conf
{
"bucket": "test1",
"tenant": "",
"zonegroup": "513b96de-2450-4292-a86a-314abbe29766",
"placement_rule": "default-placement",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
},
"id": "e34456f0-d371-4384-9007-70c60563fb0b.4281.15",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4281.15", # 由marker可以看到对应index 对象 .dir.e34456f0-d371-4384-9007-70c60563fb0b.4281.15
"index_type": "Normal",
"owner": "user1",
"ver": "0#3",
"master_ver": "0#0",
"mtime": "2019-11-20 07:53:38.898208Z",
"max_marker": "0#",
"usage": {
"rgw.main": {
"size": 10279,
"size_actual": 16384,
"size_utilized": 10279,
"size_kb": 11,
"size_kb_actual": 16,
"size_kb_utilized": 11,
"num_objects": 2
}
},
"bucket_quota": {
"enabled": false,
"check_on_raw": false,
"max_size": -1,
"max_size_kb": 0,
"max_objects": -1
}
}listomapkeys 看下
[root@stor14 build]# bin/rados -c ceph.conf listomapkeys .dir.e34456f0-d371-4384-9007-70c60563fb0b.4281.15 -p default.rgw.buckets.index
dcj1
dcj2
可以看到omap key：dcj1 和dcj2，再listomapvals 看下，此时可以看2个RGW 对象
[root@stor14 build]# bin/rados -c ceph.conf listomapvals .dir.e34456f0-d371-4384-9007-70c60563fb0b.4281.15 -p default.rgw.buckets.index
dcj1
value (218 bytes) :
00000000 08 03 d4 00 00 00 04 00 00 00 64 63 6a 31 01 00 |..........dcj1..|
00000010 00 00 00 00 00 00 01 07 03 6e 00 00 00 01 ca 13 |.........n......|
00000020 00 00 00 00 00 00 06 f1 d4 5d f1 9c e0 0e 20 00 |.........].... .|
00000030 00 00 38 30 64 34 36 33 66 36 32 35 65 39 32 37 |..80d463f625e927|
00000040 39 31 39 35 65 30 37 34 65 35 34 65 65 63 64 31 |9195e074e54eecd1|
00000050 34 39 05 00 00 00 75 73 65 72 31 05 00 00 00 75 |49....user1....u|
00000060 73 65 72 31 0a 00 00 00 74 65 78 74 2f 70 6c 61 |ser1....text/pla|
00000070 69 6e ca 13 00 00 00 00 00 00 00 00 00 00 08 00 |in..............|
00000080 00 00 53 54 41 4e 44 41 52 44 00 00 00 00 00 00 |..STANDARD......|
00000090 00 00 00 01 01 02 00 00 00 08 01 01 2c 00 00 00 |............,...|
000000a0 65 33 34 34 35 36 66 30 2d 64 33 37 31 2d 34 33 |e34456f0-d371-43|
000000b0 38 34 2d 39 30 30 37 2d 37 30 63 36 30 35 36 33 |84-9007-70c60563|
000000c0 66 62 30 62 2e 34 32 36 37 2e 35 34 00 00 00 00 |fb0b.4267.54....|
000000d0 00 00 00 00 00 00 00 00 00 00 |..........|
000000da
dcj2
value (220 bytes) :
00000000 08 03 d6 00 00 00 04 00 00 00 64 63 6a 32 01 00 |..........dcj2..|
00000010 00 00 00 00 00 00 01 07 03 6e 00 00 00 01 5d 14 |.........n....].|
00000020 00 00 00 00 00 00 54 2e d5 5d 49 86 f8 2a 20 00 |......T..]I..* .|
00000030 00 00 38 39 36 39 39 37 63 37 63 37 62 64 64 61 |..896997c7c7bdda|
00000040 30 38 61 62 30 39 62 38 35 34 36 32 33 61 30 30 |08ab09b854623a00|
00000050 30 30 05 00 00 00 75 73 65 72 31 05 00 00 00 75 |00....user1....u|
00000060 73 65 72 31 0a 00 00 00 74 65 78 74 2f 70 6c 61 |ser1....text/pla|
00000070 69 6e 5d 14 00 00 00 00 00 00 00 00 00 00 08 00 |in].............|
00000080 00 00 53 54 41 4e 44 41 52 44 00 00 00 00 00 00 |..STANDARD......|
00000090 00 00 00 01 01 02 00 00 00 08 01 02 2e 00 00 00 |................|
000000a0 65 33 34 34 35 36 66 30 2d 64 33 37 31 2d 34 33 |e34456f0-d371-43|
000000b0 38 34 2d 39 30 30 37 2d 37 30 63 36 30 35 36 33 |84-9007-70c60563|
000000c0 66 62 30 62 2e 34 35 36 34 2e 34 30 30 36 00 00 |fb0b.4564.4006..|
000000d0 00 00 00 00 00 00 00 00 00 00 00 00 |............|
000000dc一般一个存储桶对应一个rados object，H版之后加入shard 后，也可能会一个存储桶对应多个rados objects。目前N 版已经可以做到auto reshard。我们通过配置rgw_override_bucket_index_max_shards = 5 将新建存储桶的shard 初始化为5 分片
新建存储桶test2
[root@stor14 build]# s3cmd mb s3://test2
Bucket 's3://test2/' created
可以看到max_marker 有5个：0#,1#,2#,3#,4#
[root@stor14 build]# bin/radosgw-admin bucket stats --bucket test2 -c ceph.conf
{
"bucket": "test2",
"tenant": "",
"zonegroup": "513b96de-2450-4292-a86a-314abbe29766",
"placement_rule": "default-placement",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
},
"id": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"index_type": "Normal",
"owner": "user1",
"ver": "0#1,1#1,2#1,3#1,4#1",
"master_ver": "0#0,1#0,2#0,3#0,4#0",
"mtime": "2019-11-20 12:26:38.164727Z",
"max_marker": "0#,1#,2#,3#,4#",
"usage": {},
"bucket_quota": {
"enabled": false,
"check_on_raw": false,
"max_size": -1,
"max_size_kb": 0,
"max_objects": -1
}
}此时除了之前的2个 index objects 外，多出来了5个 对象，名为 .dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.
[root@stor14 build]# bin/rados -c ceph.conf ls -p default.rgw.buckets.index
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.1
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.3
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.2
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.4
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4281.15
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.0
.dir.e34456f0-d371-4384-9007-70c60563fb0b.4281.17我们上传一个对象jcd1 至新建的存储桶test2
[root@stor14 build]# s3cmd ls s3://test2
[root@stor14 build]# s3cmd put ./ceph.conf s3://test2/jcd1
upload: './ceph.conf' -> 's3://test2/jcd1' [1 of 1]
5255 of 5255 100% in 0s 26.82 kB/s done
[root@stor14 build]# bin/rados -c ceph.conf listomapkeys .dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.0 -p default.rgw.buckets.index
[root@stor14 build]# bin/rados -c ceph.conf listomapkeys .dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.1 -p default.rgw.buckets.index
[root@stor14 build]# bin/rados -c ceph.conf listomapkeys .dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.2 -p default.rgw.buckets.index
jcd1
[root@stor14 build]# bin/rados -c ceph.conf listomapkeys .dir.e34456f0-d371-4384-9007-70c60563fb0b.4848.1.3 -p default.rgw.buckets.index可以看到上传的对象在分片2 中。
默认情况下bucket index obj 并没有attr 信息：
[root@stor14 build]# bin/rados -c ceph.conf listxattr .dir.e34456f0-d371-4384-9007-70c60563fb0b.4281.15 -p default.rgw.buckets.index一般来说单shard 保存10万-15万对象为宜，shard 数也不是越多越好，过多的shard会导致部分类似list bucket的操作消耗大量底层存储IO，导致部分请求耗时过长。
data
每个RGW object 会保存在一个或多个rados obj 中。下面会详细介绍RGW Object 的layout。
RGW Pools
RGW 使用各种 pool 来专门存放特定类型的数据, zone 所使用的存储池的命名 规则 {zone}.rgw.{functions} 。
可以看到default zone下配置的存储池：
[root@stor14 build]# bin/radosgw-admin zone get -c ceph.conf
{
"id": "e34456f0-d371-4384-9007-70c60563fb0b",
"name": "default",
"domain_root": "default.rgw.meta:root",
"control_pool": "default.rgw.control",
"gc_pool": "default.rgw.log:gc",
"lc_pool": "default.rgw.log:lc",
"bl_pool": "default.rgw.log:bl",
"log_pool": "default.rgw.log",
"intent_log_pool": "default.rgw.log:intent",
"usage_log_pool": "default.rgw.log:usage",
"reshard_pool": "default.rgw.log:reshard",
"user_keys_pool": "default.rgw.meta:users.keys",
"user_email_pool": "default.rgw.meta:users.email",
"user_swift_pool": "default.rgw.meta:users.swift",
"user_uid_pool": "default.rgw.meta:users.uid",
"otp_pool": "default.rgw.otp", # metadata 中提到的One-Time Password 用的存储池
"system_key": {
"access_key": "ms_sync",
"secret_key": "ms_sync"
},
"bl_deliver_key": {
"access_key": "bl_deliver",
"secret_key": "bl_deliver"
},
"placement_pools": [
{
"key": "default-placement",
"val": {
"index_pool": "default.rgw.buckets.index",
"storage_classes": {
"STANDARD": {
"data_pool": "default.rgw.buckets.data"
}
},
"data_extra_pool": "default.rgw.buckets.non-ec",
"index_type": 0
}
}
],
"metadata_heap": "",
"realm_id": "5f47bd7c-94a1-48f5-8c51-5bb8f7075425"
}重要的RGW Pools：
.rgw.root - 保存 zone,zg,realm,period 相关的信息
zone.rgw.control - librados 提供的 watch-notify 机制保证缓存一致性,如 notify.
- zone.rgw.log - 记录不同的日志,--namespace usage, gc, lc, bl
- gc: 用于垃圾回收的
- lc: 用于lifecycle
- bl：用于bucket logging
zone.rgw.buckets.data ,数据池
zone.rgw.buckets.extra ,现在叫 rgw.buckets.non-ec ,multipart upload 过程中的临时数据会放这里。
zone.rgw.buckets.index - 存储 bucket 中对象的索引对象： “.dir.
zone.rgw.meta: 对象存储元数据池, --namespace users, root
root：
.bucket.meta. : # see put_bucket_instance_info() - users.id：Contains both per-user information (RGWUserInfo) in “
” objects and per-user lists of buckets in omaps of “ .buckets” objects. The “ ” may contain the tenant if non-empty - users.keys：允许RGW在认证时通过access-key 去找到user id
看下.rgw.meta 池
[root@stor14 build]# bin/rados -c ceph.conf ls -p default.rgw.meta -N root test1 .bucket.meta.test2:e34456f0-d371-4384-9007-70c60563fb0b.4848.1 .bucket.meta.bltest1:e34456f0-d371-4384-9007-70c60563fb0b.4281.17 bltest1 .bucket.meta.test1:e34456f0-d371-4384-9007-70c60563fb0b.4281.15 test2 [root@stor14 build]# bin/rados -c ceph.conf listxattr test1 -p default.rgw.meta -N root ceph.objclass.version [root@stor14 build]# bin/rados -c ceph.conf listxattr .bucket.meta.test1:e34456f0-d371-4384-9007-70c60563fb0b.4281.15 -p default.rgw.meta -N root ceph.objclass.version user.rgw.acl
看下users.uid
[root@stor14 build]# bin/rados -c ceph.conf ls -p default.rgw.meta -N users.uid
56789abcdef0123456789abcdef0123456789abcdef0123456789abcdef01234
bl_deliver
testx$9876543210abcdef0123456789abcdef0123456789abcdef0123456789abcdef
0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef
user1
test
testid
ms_sync
user1.buckets # *****
[root@stor14 build]# bin/radosgw-admin user list -c ceph.conf
[
"56789abcdef0123456789abcdef0123456789abcdef0123456789abcdef01234",
"bl_deliver",
"testx$9876543210abcdef0123456789abcdef0123456789abcdef0123456789abcdef",
"0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef",
"user1",
"test",
"testid",
"ms_sync"
]可以看到users.uid 中除了oss user，还有一个user1.buckets
“Contains both per-user information (RGWUserInfo) in “
其实这个user1.buckets 对象是一个非常重要的对象。通过bucket index obj 可以list 某个存储桶中的rgw objects，如何list 某个对象存储用户下的存储桶呢？答案就是这个
[root@stor14 build]# bin/rados -c ceph.conf listomapkeys user1.buckets -p default.rgw.meta -N users.uid
bltest1
test1
test2
[root@stor14 build]# bin/rados -c ceph.conf listomapvals user1.buckets -p default.rgw.meta -N users.uid
bltest1
value (172 bytes) :
00000000 09 05 a6 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
00000010 00 00 03 f1 d4 5d 00 00 00 00 00 00 00 00 07 03 |.....]..........|
00000020 77 00 00 00 07 00 00 00 62 6c 74 65 73 74 31 00 |w.......bltest1.|
00000030 00 00 00 2c 00 00 00 65 33 34 34 35 36 66 30 2d |...,...e34456f0-|
00000040 64 33 37 31 2d 34 33 38 34 2d 39 30 30 37 2d 37 |d371-4384-9007-7|
00000050 30 63 36 30 35 36 33 66 62 30 62 2e 34 32 38 31 |0c60563fb0b.4281|
00000060 2e 31 37 2c 00 00 00 65 33 34 34 35 36 66 30 2d |.17,...e34456f0-|
00000070 64 33 37 31 2d 34 33 38 34 2d 39 30 30 37 2d 37 |d371-4384-9007-7|
00000080 30 63 36 30 35 36 33 66 62 30 62 2e 34 32 38 31 |0c60563fb0b.4281|
00000090 2e 31 37 00 00 00 00 00 00 00 00 00 00 00 00 00 |.17.............|
000000a0 00 00 00 01 03 f1 d4 5d 26 99 cc 17 |.......]&...|
000000ac
test1
value (170 bytes) :
00000000 09 05 a4 00 00 00 00 00 00 00 27 28 00 00 00 00 |..........'(....|
00000010 00 00 02 f1 d4 5d 02 00 00 00 00 00 00 00 07 03 |.....]..........|
00000020 75 00 00 00 05 00 00 00 74 65 73 74 31 00 00 00 |u.......test1...|
00000030 00 2c 00 00 00 65 33 34 34 35 36 66 30 2d 64 33 |.,...e34456f0-d3|
00000040 37 31 2d 34 33 38 34 2d 39 30 30 37 2d 37 30 63 |71-4384-9007-70c|
00000050 36 30 35 36 33 66 62 30 62 2e 34 32 38 31 2e 31 |60563fb0b.4281.1|
00000060 35 2c 00 00 00 65 33 34 34 35 36 66 30 2d 64 33 |5,...e34456f0-d3|
00000070 37 31 2d 34 33 38 34 2d 39 30 30 37 2d 37 30 63 |71-4384-9007-70c|
00000080 36 30 35 36 33 66 62 30 62 2e 34 32 38 31 2e 31 |60563fb0b.4281.1|
00000090 35 00 00 00 00 00 00 00 00 00 40 00 00 00 00 00 |5.........@.....|
000000a0 00 01 02 f1 d4 5d 74 a9 42 32 |.....]t.B2|
000000aa
test2
value (168 bytes) :
00000000 09 05 a2 00 00 00 00 00 00 00 87 14 20 02 00 00 |............ ...|
00000010 00 00 fd 30 d5 5d 05 00 00 00 00 00 00 00 07 03 |...0.]..........|
00000020 73 00 00 00 05 00 00 00 74 65 73 74 32 00 00 00 |s.......test2...|
00000030 00 2b 00 00 00 65 33 34 34 35 36 66 30 2d 64 33 |.+...e34456f0-d3|
00000040 37 31 2d 34 33 38 34 2d 39 30 30 37 2d 37 30 63 |71-4384-9007-70c|
00000050 36 30 35 36 33 66 62 30 62 2e 34 38 34 38 2e 31 |60563fb0b.4848.1|
00000060 2b 00 00 00 65 33 34 34 35 36 66 30 2d 64 33 37 |+...e34456f0-d37|
00000070 31 2d 34 33 38 34 2d 39 30 30 37 2d 37 30 63 36 |1-4384-9007-70c6|
00000080 30 35 36 33 66 62 30 62 2e 34 38 34 38 2e 31 00 |0563fb0b.4848.1.|
00000090 00 00 00 00 00 00 00 00 20 20 02 00 00 00 00 01 |........ ......|
000000a0 fd 30 d5 5d 57 b6 f2 3a |.0.]W..:|
000000a8RGW Object
一个RGW Object 有一个或多个Rados Object 组成。RGW object 逻辑上一般分为head 和tail。
其中head( object logiccal head, olh)，元数据会保存在扩展属性xattr中:
- head 保存在单个 RADOS object
- head 是有固定的大小的(rgw_max_chunk_size 默认 4M)
- 内容可变 保存 RGW object 的元数据
- ACLs
- Manifest
- Conten-Type
- ETag
- User-Defined Metadata(保存在 xattrs)
tail 部分:
- 内容不可变
- 对于小于 rgw_max_chunk_size 的 rgw-object 没有 tail
- 对于multipart 上传的对象，tail 分multipart （每个part 的第一个rados对象）和shadow
- tail 被分隔成若干个 part，单个 part 保存 rgw-object 的连续数据,每个 part 默认 15M
- 普通的 rgw-object 只有单个 part(但 part 内可以有很多的 stripe),对于 multipart - 上传的 rgw-object 就有多个 part(最后一个 part 可能比较小)。 注意 part 还是逻辑上的
- 每个 part 都是 striped, striped part 由默认的 stripe size,每个 stripe part 映射到 rados-object,stripe 和 rados-object 一一对应
验证一下：
[root@stor14 build]# s3cmd put ./4MB s3://test2
upload: './4MB' -> 's3://test2/4MB' [1 of 1]
4194304 of 4194304 100% in 0s 8.14 MB/s done
[root@stor14 build]# bin/rados -c ceph.conf ls -p default.rgw.buckets.data
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_jcd1
e34456f0-d371-4384-9007-70c60563fb0b.4281.15_dcj2
e34456f0-d371-4384-9007-70c60563fb0b.4281.15_dcj1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_4MB
[root@stor14 build]# dd if=/dev/zero of=./5MB count=5 bs=1024k
记录了5+0 的读入
记录了5+0 的写出
5242880字节(5.2 MB)已复制，0.00357833 秒，1.5 GB/秒
[root@stor14 build]# s3cmd put ./5MB s3://test2
upload: './5MB' -> 's3://test2/5MB' [1 of 1]
5242880 of 5242880 100% in 0s 7.18 MB/s done
[root@stor14 build]# bin/rados -c ceph.conf ls -p default.rgw.buckets.data
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_jcd1
e34456f0-d371-4384-9007-70c60563fb0b.4281.15_dcj2
e34456f0-d371-4384-9007-70c60563fb0b.4281.15_dcj1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_4MB
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_5MB
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_.lWnArC6BcKPkRjbGf-B0q23BSpxjql1_1我们先后有上传了2个对象:一个4MB，一个5MB，大小同对象名。可以看到，超过4MB 之后，RGW 对象分为了2 个RADOS 对象，其中
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_5MB 为head，e34456f0-d371-4384-9007-70c60563fb0b.4848.1_shadow.lWnArC6BcKPkRjbGf-B0q23BSpxjql1_1 为tail。
查看2个rados obj 的大小：
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1_5MB
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1_5MB mtime 2019-11-20 20:49:11.000000, size 4194304
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_.lWnArC6BcKPkRjbGf-B0q23BSpxjql1_1
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_.lWnArC6BcKPkRjbGf-B0q23BSpxjql1_1 mtime 2019-11-20 20:49:11.000000, size 1048576可以看到，head obj size=4194304，也即一个stripe 大小，加上tail obj size=1048576，正好等于RGW raw obj 的大小5242880。
我们可以在head 中可以list 元数据
[root@stor14 build]# bin/rados -c ceph.conf listxattr e34456f0-d371-4384-9007-70c60563fb0b.4848.1_5MB -p default.rgw.buckets.data
user.rgw.acl
user.rgw.content_type
user.rgw.etag
user.rgw.idtag
user.rgw.manifest # manifest 保存了RGW 对象的layout，可以看作是RGW 对象到RADOS 对象的映射关系图。
user.rgw.pg_ver
user.rgw.source_zone
user.rgw.storage_class
user.rgw.tail_tag
user.rgw.x-amz-content-sha256
user.rgw.x-amz-date
user.rgw.x-amz-meta-s3cmd-attrs看下manifest
[root@stor14 build]# bin/rados -c ceph.conf getxattr e34456f0-d371-4384-9007-70c60563fb0b.4848.1_5MB -p default.rgw.buckets.data user.rgw.manifest > /tmp/mdecode 为可读：
[root@stor14 build]# bin/ceph-dencoder type RGWObjManifest import /tmp/m decode dump_json
{
"objs": [],
"obj_size": 5242880,
"explicit_objs": "false",
"head_size": 4194304, # head 大小
"max_head_size": 4194304,
"prefix": ".lWnArC6BcKPkRjbGf-B0q23BSpxjql1_", # 可以在tail obj 上看到，关联各tail objects
"rules": [
{
"key": 0,
"val": {
"start_part_num": 0,
"start_ofs": 4194304,
"part_size": 0,
"stripe_max_size": 4194304,
"override_prefix": ""
}
}
],
"tail_instance": "",
"tail_placement": {
"bucket": {
"name": "test2",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"bucket_id": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"tenant": "",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
}
},
"placement_rule": "default-placement"
},
"begin_iter": {
"part_ofs": 0,
"stripe_ofs": 0,
"ofs": 0,
"stripe_size": 4194304,
"cur_part_id": 0,
"cur_stripe": 0,
"cur_override_prefix": "",
"location": {
"placement_rule": "default-placement",
"obj": {
"bucket": {
"name": "test2",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"bucket_id": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"tenant": "",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
}
},
"key": {
"name": "5MB",
"instance": "",
"ns": ""
}
},
"raw_obj": {
"pool": "",
"oid": "",
"loc": ""
},
"is_raw": false
}
},
"end_iter": {
"part_ofs": 4194304,
"stripe_ofs": 4194304,
"ofs": 5242880,
"stripe_size": 1048576,
"cur_part_id": 0,
"cur_stripe": 1,
"cur_override_prefix": "",
"location": {
"placement_rule": "default-placement",
"obj": {
"bucket": {
"name": "test2",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"bucket_id": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"tenant": "",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
}
},
"key": {
"name": ".lWnArC6BcKPkRjbGf-B0q23BSpxjql1_1", # 当前shadow tail obj的key：manifest.prefix+stripe_id
"instance": "",
"ns": "shadow" # tail 类型为shadow
}
},
"raw_obj": {
"pool": "",
"oid": "",
"loc": ""
},
"is_raw": false
}
}
}manifest 中begin_iter 和end_iter的ofset 值及cur_part_id、cur_stripe_id，这些数据决定了RGW 在RADOS层的数据布局：当前5MB的RGW obj，不分part，分为2个stripe，第一个stripe 4194304（4MB），剩余1048576在第二个stripe，结束ofs为5242880。rules 中start_part_num=0 表示当前未分part。
看一个分part 上传的manifest:
root@stor14 build]# bin/rados -c ceph.conf getxattr e34456f0-d371-4384-9007-70c60563fb0b.4848.1_20MB -p default.rgw.buckets.data user.rgw.manifest > /tmp/m
[root@stor14 build]# bin/ceph-dencoder type RGWObjManifest import /tmp/m decode dump_json
{
"objs": [],
"obj_size": 20971520,
"explicit_objs": "false",
"head_size": 0,
"max_head_size": 0,
"prefix": "20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx",
"rules": [
{
"key": 0,
"val": {
"start_part_num": 1, # "When uploading a part, in addition to the upload ID, you must specify a part number. You can choose any part number between 1 and 10,000. A part number uniquely identifies a part and its position in the object you are uploading. "
"start_ofs": 0,
"part_size": 15728640,
"stripe_max_size": 4194304,
"override_prefix": ""
}
},
{
"key": 15728640,
"val": {
"start_part_num": 2,
"start_ofs": 15728640,
"part_size": 5242880,
"stripe_max_size": 4194304,
"override_prefix": ""
}
}
],
"tail_instance": "",
"tail_placement": {
"bucket": {
"name": "test2",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"bucket_id": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"tenant": "",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
}
},
"placement_rule": "default-placement"
},
"begin_iter": {
"part_ofs": 0,
"stripe_ofs": 0,
"ofs": 0,
"stripe_size": 4194304,
"cur_part_id": 1,
"cur_stripe": 0,
"cur_override_prefix": "",
"location": {
"placement_rule": "default-placement",
"obj": {
"bucket": {
"name": "test2",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"bucket_id": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"tenant": "",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
}
},
"key": {
"name": "20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1", # first part key
"instance": "",
"ns": "multipart" # tail 类型为multipart
}
},
"raw_obj": {
"pool": "",
"oid": "",
"loc": ""
},
"is_raw": false
}
},
"end_iter": {
"part_ofs": 20971520,
"stripe_ofs": 20971520,
"ofs": 20971520,
"stripe_size": 4194304,
"cur_part_id": 3, ###??
"cur_stripe": 0,
"cur_override_prefix": "",
"location": {
"placement_rule": "default-placement",
"obj": {
"bucket": {
"name": "test2",
"marker": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"bucket_id": "e34456f0-d371-4384-9007-70c60563fb0b.4848.1",
"tenant": "",
"explicit_placement": {
"data_pool": "",
"data_extra_pool": "",
"index_pool": ""
}
},
"key": {
"name": "20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.3", # last part key ,3 ??
"instance": "",
"ns": "multipart" # tail 类型为multipart
}
},
"raw_obj": {
"pool": "",
"oid": "",
"loc": ""
},
"is_raw": false
}
}
}20MB大小的对象分为2个part 上传。观察manifest.rules，当前
key=0 part，也即第一个part，其start_part_num=1，part_size=15728640；
key=1 part，也即第二个part，其start_part_num=2，part_size=5242880，也即剩余大小；
AWS S3 要求multipart 时part number需要在1-1000间，不可以使用0。
ACL
[root@stor14 build]# bin/rados -c ceph.conf getxattr e34456f0-d371-4384-9007-70c60563fb0b.4848.1_5MB -p default.rgw.buckets.data user.rgw.acl > /tmp/m
[root@stor14 build]# bin/ceph-dencoder type RGWAccessControlPolicy import /tmp/m decode dump_json
{
"acl": {
"acl_user_map": [
{
"user": "user1",
"acl": 15
}
],
"acl_group_map": [],
"grant_map": [
{
"id": "user1",
"grant": {
"type": {
"type": 0
},
"id": "user1",
"email": "",
"permission": {
"flags": 15
},
"name": "user1",
"group": 0,
"url_spec": ""
}
}
]
},
"owner": {
"id": "user1",
"display_name": "user1"
}
}如果RGW 对象大小超过15MB, 利用s3cmd 上传会自动分段上传，也即multipart 上传。
[root@stor14 build]# s3cmd put ./20MB s3://test2
upload: './20MB' -> 's3://test2/20MB' [part 1 of 2, 15MB] [1 of 1]
15728640 of 15728640 100% in 1s 11.65 MB/s done
upload: './20MB' -> 's3://test2/20MB' [part 2 of 2, 5MB] [1 of 1]
5242880 of 5242880 100% in 0s 8.21 MB/s done
[root@stor14 build]# bin/rados -c ceph.conf ls -p default.rgw.buckets.data
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_2
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_jcd1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2_1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_1
e34456f0-d371-4384-9007-70c60563fb0b.4281.15_dcj2
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_20MB
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_3
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2
e34456f0-d371-4384-9007-70c60563fb0b.4281.15_dcj1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_4MB
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_5MB
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_.lWnArC6BcKPkRjbGf-B0q23BSpxjql1_1
[root@stor14 build]# bin/rados -c ceph.conf ls -p default.rgw.buckets.data|grep 20MB
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_2
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2_1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1_20MB
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_3
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2这个20MB 的RGW 对象分成了7 个RADOS 对象，其中
- head object：e34456f0-d371-4384-9007-70c60563fb0b.4848.1_20MB
- tail objects：
- multipart obj：e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1
- shadow objs：e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_1,
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_2,
e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_3
- shadow objs：e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_1,
- multipart obj：e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2
- shadow objs：e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2_1
- multipart obj：e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1
可以看到，multipart 上传的对象会有一个head，默认每个part 15M，每个part 的第一个tail 为multipart obj，剩余都为shadow obj。
查看个rados 的大小：
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1_20MB
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1_20MB mtime 2019-11-20 20:58:01.000000, size 0
head obj size 为0，multipart 情况head 只存元数据xattr
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1 mtime 2019-11-20 20:58:00.000000, size 4194304
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_1
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_1 mtime 2019-11-20 20:58:00.000000, size 4194304
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_2
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_2 mtime 2019-11-20 20:58:00.000000, size 4194304
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_3
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_3 mtime 2019-11-20 20:58:00.000000, size 3145728
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2 mtime 2019-11-20 20:58:01.000000, size 4194304
[root@stor14 build]# bin/rados -c ceph.conf -p default.rgw.buckets.data stat e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2_1
default.rgw.buckets.data/e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2_1 mtime 2019-11-20 20:58:01.000000, size 10485760 + 4194304 + 4194304 + 4194304 + 3145728 + 4194304 + 1048576 = 20971520，正好是RGW obj "20MB"的大小。
一般的tail obj（shadow obj）中没有元数据信息，不过multipart obj 会有一些元数据信息：
[root@stor14 build]# bin/rados -c ceph.conf listxattr e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_2 -p default.rgw.buckets.data
[root@stor14 build]# bin/rados -c ceph.conf listxattr e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1 -p default.rgw.buckets.data
user.rgw.acl
user.rgw.etag
user.rgw.pg_ver
user.rgw.source_zone
user.rgw.x-amz-content-sha256
user.rgw.x-amz-date可以看到没有manifest。
我们可以从multipart tail obj 中拿到和从head obj中一样的acl 元数据。
[root@stor14 build]# bin/rados -c ceph.conf getxattr e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.2 -p default.rgw.buckets.data user.rgw.acl > /tmp/m
[root@stor14 build]# bin/ceph-dencoder type RGWAccessControlPolicy import /tmp/m decode dump_json
{
"acl": {
"acl_user_map": [
{
"user": "user1",
"acl": 15
}
],
"acl_group_map": [],
"grant_map": [
{
"id": "user1",
"grant": {
"type": {
"type": 0
},
"id": "user1",
"email": "",
"permission": {
"flags": 15
},
"name": "user1",
"group": 0,
"url_spec": ""
}
}
]
},
"owner": {
"id": "user1",
"display_name": "user1"
}
}在未开启versioning的情况下，RGW obj 对应的RADOS obj 的命名规则为
head：
__ ，如 e34456f0-d371-4384-9007-70c60563fb0b.4848.1_20MB - （multipart上传）mulpart tail:
__multipart_ . . ，如 e34456f0-d371-4384-9007-70c60563fb0b.4848.1__multipart_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1 - shadow tail：
- multipart 上传：
__shadow_ . . _ ，如 e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_20MB.2~PDSpB4F7hIgKhjVwf6NqmJtBuKxm-mx.1_3 - 非multipart 上传：
__shadow_. _ ，如 e34456f0-d371-4384-9007-70c60563fb0b.4848.1__shadow_.lWnArC6BcKPkRjbGf-B0q23BSpxjql1_1
- multipart 上传：
其中marker 就是bucket stats 中看到的marker，也是bucket id。
在开启versioning 之后，会增加一个object_instance_id，这里暂不做详细讨论。
References
- https://docs.ceph.com/docs/master/radosgw/layout/
- https://docs.aws.amazon.com/IAM/latest/UserGuide/id_credentials_mfa.html
- https://www.authonet.com/authentication.php
- https://docs.aws.amazon.com/AmazonS3/latest/dev/mpuoverview.html