describe architecture
Change-Id: Ifc2cdddda95d0bfead3b666dd3625c6bbcc41fce
diff --git a/README.rst b/README.rst
index c7db5ea..d5e1c98 100644
--- a/README.rst
+++ b/README.rst
@@ -12,6 +12,167 @@
Use salt-formula-linux for initial disk partitioning.
+Daemons
+--------
+
+Ceph uses several daemons to handle data and cluster state. Each daemon type requires different computing capacity and hardware optimization.
+
+These daemons are currently supported by formula:
+
+* MON (`ceph.mon`)
+* OSD (`ceph.osd`)
+* RGW (`ceph.radosgw`)
+
+
+Architecture decisions
+-----------------------
+
+Please refer to upstream achritecture documents before designing your cluster. Solid understanding of Ceph principles is essential for making architecture decisions described bellow.
+http://docs.ceph.com/docs/master/architecture/
+
+* Ceph version
+
+There is 3 or 4 stable releases every year and many of nighty/dev release. You should decide which version will be used since the only stable releases are recommended for production. Some of the releases are marked LTS (Long Term Stable) and these releases receive bugfixed for longer period - usually until next LTS version is released.
+
+* Number of MON daemons
+
+Use 1 MON daemon for testing, 3 MONs for smaller production clusters and 5 MONs for very large production cluster. There is no need to have more than 5 MONs in normal environment because there isn't any significant benefit in running more than 5 MONs. Ceph require MONS to form quorum so you need to heve more than 50% of the MONs up and running to have fully operational cluster. Every I/O operation will stop once less than 50% MONs is availabe because they can't form quorum.
+
+* Number of PGs
+
+Placement groups are providing mappping between stored data and OSDs. It is necessary to calculate number of PGs because there should be stored decent amount of PGs on each OSD. Please keep in mind *decreasing number of PGs* isn't possible and *increading* can affect cluster performance.
+
+http://docs.ceph.com/docs/master/rados/operations/placement-groups/
+http://ceph.com/pgcalc/
+
+* Daemon colocation
+
+It is recommended to dedicate nodes for MONs and RWG since colocation can have and influence on cluster operations. Howerver, small clusters can be running MONs on OSD node but it is critical to have enough of resources for MON daemons because they are the most important part of the cluster.
+
+Installing RGW on node with other daemons isn't recommended because RGW daemon usually require a lot of bandwith and it harm cluster health.
+
+* Journal location
+
+There are two way to setup journal:
+ * **Colocated** journal is located (usually at the beginning) on the same disk as partition for the data. This setup is easier for installation and it doesn't require any other disk to be used. However, colocated setup is significantly slower than dedicated)
+ * **Dedicate** journal is placed on different disk than data. This setup can deliver much higher performance than colocated but it require to have more disks in servers. Journal drives should be carefully selected because high I/O and durability is required.
+
+* Store type (Bluestore/Filestore)
+
+Recent version of Ceph support Bluestore as storage backend and backend should be used if available.
+
+http://docs.ceph.com/docs/master/rados/configuration/bluestore-config-ref/
+
+* Cluster and public network
+
+Ceph cluster is accessed using network and thus you need to have decend capacity to handle all the client. There are two networks required for cluster: **public** network and cluster network. Public network is used for client connections and MONs and OSDs are listening on this network. Second network ic called **cluster** networks and this network is used for communication between OSDs.
+
+Both networks should have dedicated interfaces, bonding interfaces and dedicating vlans on bonded interfaces isn't allowed. Good practise is dedicate more throughput for the cluster network because cluster traffic is more important than client traffic.
+
+* Pool parameters (size, min_size, type)
+
+You should setup each pool according to it's expected usage, at least `min_size` and `size` and pool type should be considered.
+
+* Cluster monitoring
+
+* Hardware
+
+Please refer to upstream hardware recommendation guide for general information about hardware.
+
+Ceph servers are required to fulfil special requirements becauce load generated by Ceph can be diametrically opposed to common load.
+
+http://docs.ceph.com/docs/master/start/hardware-recommendations/
+
+
+Basic management commands
+------------------------------
+
+Cluster
+********
+
+- :code:`ceph health` - check if cluster is healthy (:code:`ceph health detail` can provide more information)
+
+
+.. code-block:: bash
+
+ root@c-01:~# ceph health
+ HEALTH_OK
+
+- :code:`ceph status` - shows basic information about cluster
+
+
+.. code-block:: bash
+
+ root@c-01:~# ceph status
+ cluster e2dc51ae-c5e4-48f0-afc1-9e9e97dfd650
+ health HEALTH_OK
+ monmap e1: 3 mons at {1=192.168.31.201:6789/0,2=192.168.31.202:6789/0,3=192.168.31.203:6789/0}
+ election epoch 38, quorum 0,1,2 1,2,3
+ osdmap e226: 6 osds: 6 up, 6 in
+ pgmap v27916: 400 pgs, 2 pools, 21233 MB data, 5315 objects
+ 121 GB used, 10924 GB / 11058 GB avail
+ 400 active+clean
+ client io 481 kB/s rd, 132 kB/s wr, 185 op/
+
+MON
+****
+
+http://ceph.com/docs/master/rados/troubleshooting/troubleshooting-mon/
+
+OSD
+****
+
+http://ceph.com/docs/master/rados/troubleshooting/troubleshooting-osd/
+
+- :code:`ceph osd tree` - show all OSDs and it's state
+
+.. code-block:: bash
+
+ root@c-01:~# ceph osd tree
+ ID WEIGHT TYPE NAME UP/DOWN REWEIGHT PRIMARY-AFFINITY
+ -4 0 host c-04
+ -1 10.79993 root default
+ -2 3.59998 host c-01
+ 0 1.79999 osd.0 up 1.00000 1.00000
+ 1 1.79999 osd.1 up 1.00000 1.00000
+ -3 3.59998 host c-02
+ 2 1.79999 osd.2 up 1.00000 1.00000
+ 3 1.79999 osd.3 up 1.00000 1.00000
+ -5 3.59998 host c-03
+ 4 1.79999 osd.4 up 1.00000 1.00000
+ 5 1.79999 osd.5 up 1.00000 1.00000
+
+- :code:`ceph osd pools ls` - list of pool
+
+.. code-block:: bash
+
+ root@c-01:~# ceph osd lspools
+ 0 rbd,1 test
+
+PG
+***
+
+http://ceph.com/docs/master/rados/troubleshooting/troubleshooting-pg
+
+- :code:`ceph pg ls` - list placement groups
+
+.. code-block:: bash
+
+ root@c-01:~# ceph pg ls | head -n 4
+ pg_stat objects mip degr misp unf bytes log disklog state state_stamp v reported up up_primary acting acting_primary last_scrub scrub_stamp last_deep_scrub deep_scrub_stamp
+ 0.0 11 0 0 0 0 46137344 3044 3044 active+clean 2015-07-02 10:12:40.603692 226'10652 226:1798 [4,2,0] 4 [4,2,0] 4 0'0 2015-07-01 18:38:33.126953 0'0 2015-07-01 18:17:01.904194
+ 0.1 7 0 0 0 0 25165936 3026 3026 active+clean 2015-07-02 10:12:40.585833 226'5808 226:1070 [2,4,1] 2 [2,4,1] 2 0'0 2015-07-01 18:38:32.352721 0'0 2015-07-01 18:17:01.904198
+ 0.2 18 0 0 0 0 75497472 3039 3039 active+clean 2015-07-02 10:12:39.569630 226'17447 226:3213 [3,1,5] 3 [3,1,5] 3 0'0 2015-07-01 18:38:34.308228 0'0 2015-07-01 18:17:01.904199
+
+- :code:`ceph pg map 1.1` - show mapping between PG and OSD
+
+.. code-block:: bash
+
+ root@c-01:~# ceph pg map 1.1
+ osdmap e226 pg 1.1 (1.1) -> up [5,1,2] acting [5,1,2]
+
+
+
Sample pillars
==============
@@ -271,7 +432,7 @@
crush_ruleset_name: 0
Generate CRUSH map
-+++++++++++++++++++
+--------------------
It is required to define the `type` for crush buckets and these types must start with `root` (top) and end with `host`. OSD daemons will be assigned to hosts according to it's hostname. Weight of the buckets will be calculated according to weight of it's childen.