This sole purpose of this project is to learn about ceph storage and figure out the best way use Ceph from Node.js

1. Background
2. Installing Ceph
3. Compiling Ceph
4. Starting Ceph
5. Stopping Ceph
6. Installing librados
7. Node N-API example
8. Node Amazon S3 SDK example
9. N-API vs Amazon S3 SDK

This section will contains information about Ceph as I don't know anything at all about it, nor am I very familiar with storage solutions either.

Ceph is a storage system that does not have a centralized metadata server and uses a an algorithm called Controlled Replication Under Scalable Hashing (CRUSH). Instead of asking a central metadata server where data is stored the location of data is calculated.

There are 3 types of storage that Ceph provides:

• Block storage via RADOS Block Device (RBD)
• Object storage via RADOS Gateway
• File storage via CephFS

This is the object store upon which everything else is built. This layer contains the Object Store Daeomons (OSD). These daemons are totally independant and form peer-to-peer relationships to form a cluster (sounds similar to what is/was possible when using jgroups). An OSD is normally mapped to a single disk (where traditional approaches would use multiple disks and a RAID controller).

Each object is a stream of binary data which is associated key when created.

A storage cluster consists of monitors, and OSD daemons.

These are responsible for providing a known cluster state including membership and use cluster maps for this. Cluster clients retrives the cluster map from one of the monitors. OSDs also report back there state to the monitors.

A monitor as a unique id this called fsid which stands for File System ID which was a little strange until I read that this is a remainder from the days when Ceph mostly a Ceph FS.

A monitor also needs to know the clustername (the default is ceph). Monitor name can be set but defaults to the hostname of the node it runs on.

A monitor map is generated using the fsid, clustername, and one or more hostname and IP addresses.

Monitor keyring for communication with other monitors. This must be provided when bootstrapping the initial monitor.

Administrator keyring which is used by the ceph CLI tools, which requires a user named client.admin. This user must also be added to the monitor keyring.

This the what actually stores the data on a file system and providing access to this data. The data that these OSDs store has an identifier, binary data, and metadata consiting of key/value pairs. This data can be accessed through the Ceph API, using Amazon S2 services, or using the provided REST based API.

Are logical groups of ceph objects.

Is a library allowing direct access to RADOS.

Is a REST-based gateway.

This is the name of the algorighm that Ceph uses which is what determines where to place data. The CRUSH map provides a view of what the cluster looks like.

Before a client can start storing/retreiving data it needs to contact one of the monitors to get the various cluster maps. With the cluster map the client is made aware of all the monitors in the cluster, all the OSDs, and the metadata daemons in the cluster. But it does not know anything about where objects are stored. The location of where data is stored is calculated/computed.

Once it has these it can operate independently as the client is cluster aware. The client writes an object to a pool and specifies an object id.

Both ceph clients and OSD daemons depend on having knowledge about the cluster. This is information is provided in five maps.

$ ceph mon dump
dumped monmap epoch 1
epoch 1
fsid 82caea9a-9fd3-11ea-b761-482ae34923e7
last_changed 2020-05-27T04:35:41.649530+0000
created 2020-05-27T04:35:41.649530+0000
min_mon_release 15 (octopus)
0: [v2:192.168.1.74:3300/0,v1:192.168.1.74:6789/0] mon.localhost.localdomain

ceph osd dump
epoch 4
fsid 82caea9a-9fd3-11ea-b761-482ae34923e7
created 2020-05-27T04:35:43.049269+0000
modified 2020-05-27T04:35:58.723837+0000
flags sortbitwise,recovery_deletes,purged_snapdirs,pglog_hardlimit
crush_version 1
full_ratio 0.95
backfillfull_ratio 0.9
nearfull_ratio 0.85
require_min_compat_client jewel
min_compat_client jewel
require_osd_release octopus
pool 1 'device_health_metrics' replicated size 3 min_size 2 crush_rule 0 object_hash rjenkins pg_num 1 pgp_num 1 autoscale_mode on last_change 3 flags hashpspool,creating stripe_width 0 pg_num_min 1 application mgr_devicehealth
max_osd 0
blacklist 192.168.1.74:0/3903093624 expires 2020-05-28T04:35:58.723811+0000
blacklist 192.168.1.74:0/3640786923 expires 2020-05-28T04:35:58.723811+0000
blacklist 192.168.1.74:6801/1724551218 expires 2020-05-28T04:35:58.723811+0000
blacklist 192.168.1.74:0/4057683 expires 2020-05-28T04:35:58.723811+0000
blacklist 192.168.1.74:6800/1724551218 expires 2020-05-28T04:35:58.723811+0000

$ ceph pg dump

$ ceph osd getcrushmap -o compressed_crush_map
1
$ crushtool -d compressed_crush_map -o decompressed-crushmap

Output of decompressed-crushmap:

# begin crush map
tunable choose_local_tries 0
tunable choose_local_fallback_tries 0
tunable choose_total_tries 50
tunable chooseleaf_descend_once 1
tunable chooseleaf_vary_r 1
tunable chooseleaf_stable 1
tunable straw_calc_version 1
tunable allowed_bucket_algs 54

# devices

# types
type 0 osd
type 1 host
type 2 chassis
type 3 rack
type 4 row
type 5 pdu
type 6 pod
type 7 room
type 8 datacenter
type 9 zone
type 10 region
type 11 root

# buckets
root default {
	id -1		# do not change unnecessarily
	# weight 0.000
	alg straw2
	hash 0	# rjenkins1
}

# rules
rule replicated_rule {
	id 0
	type replicated
	min_size 1
	max_size 10
	step take default
	step chooseleaf firstn 0 type host
	step emit
}

# end crush map

Is a cloud object protocol used with amazons online file storage web service. Is not opensource.

OpenSource and originally developed by Rackspace. A common usecase for Swift is to virtual machine images. Implemented in Python and perhaps one in Go (not clear on this yet).

This section will go through installing Ceph on Fedora. Most examples that I've come accross use a tool named ceph-deploy but this tool is no longer maintained and does not work with my version of Fedora (link).

$ curl --silent --remote-name --location https://github.com/ceph/ceph/raw/octopus/src/cephadm/cephadm
$ chmod 744 cephadm
$ sudo ./cephadm add-repo --release octopus

Currently, there is now repository for fc31 which I'm using, so /etc/yum.repos.d/ceph.repo needs to be updated to use el8 instead which I hope will work.

Next install ceph:

$ sudo ./cephadm install

This will actually install an older version of ceph as there are package dependencies that cannot be met. See commit history for notes about this (I'm trying to keep unnecessary details out of this document).

I've needed to set my hostname to not be a fully qualified domain name before beeing able to successfully bootstrap the cluster:

$ sudo hostname ip-192-168-1-74
$ sudo ./cephadm bootstrap --mon-ip 192.168.1.74
INFO:cephadm:Ceph Dashboard is now available at:

	     URL: https://ip-192-168-1-74:8443/
	    User: admin
	Password: 04t7wb1wkx

INFO:cephadm:You can access the Ceph CLI with:

	sudo ./cephadm shell --fsid f8fb78fe-a3f4-11ea-96f7-482ae34923e7 -c /etc/ceph/ceph.conf -k /etc/ceph/ceph.client.admin.keyring

INFO:cephadm:Please consider enabling telemetry to help improve Ceph:

	ceph telemetry on

For more information see:

	https://docs.ceph.com/docs/master/mgr/telemetry/

INFO:cephadm:Bootstrap complete.

Get the status of the Ceph cluster:

$ sudo ceph -s
  cluster:
    id:     62418068-9a65-11ea-b0db-482ae34923e7
    health: HEALTH_WARN
            2 stray daemons(s) not managed by cephadm
            1 stray host(s) with 2 daemon(s) not managed by cephadm
            Reduced data availability: 1 pg inactive
            OSD count 0 < osd_pool_default_size 3
 
  services:
    mon: 1 daemons, quorum localhost.localdomain (age 18m)
    mgr: localhost.localdomain.tdctad(active, since 18m)
    osd: 0 osds: 0 up, 0 in
 
  data:
    pools:   1 pools, 1 pgs
    objects: 0 objects, 0 B
    usage:   0 B used, 0 B / 0 B avail
    pgs:     100.000% pgs unknown
             1 unknown

Note that is you forget the sudo command you'll get the following error:

$ ceph -s
[errno 13] error connecting to the cluster

This will also happen with a node client if you try to connect. For now, I've just chmod:ed all the files in /etc/ceph:

$ sudo chmod 777 /etc/ceph/*

For our experimention we need to create a pool:

$ ceph osd pool create data
pool 'data' created
$ ceph osd lspools

This is the name we use in index.js.

I'm able to connect to the cluster and create an IoCtx using the pool created above, but when I try call io_ctx.write_full nothing happens, the process hangs for ever and I've not been able to find anything in the logs which could point me to the issue. I should not the I have absolutely know idea if this should work, especially with regards to the configuration of the cluster and in particular the OSD created above.

First clone ceph:

$ git clone git@github.com:ceph/ceph
$ cd ceph
$ git submodule update --init --recursive

Next, we use cmake to generate make files:

$ mkdir build
$ cd build
$ cmake ..
$ make -j8

After having built ceph run the following command from the checked out ceph repository:

$ cd build
$ env MON=1 OSD=1 MDS=1 MGR=1 RGW=1 ../src/vstart.sh -n -d --localhost
...

setting up user testid
setting up s3-test users
setting up user tester

S3 User Info:
  access key:  0555b35654ad1656d804
  secret key:  h7GhxuBLTrlhVUyxSPUKUV8r/2EI4ngqJxD7iBdBYLhwluN30JaT3Q==

Swift User Info:
  account   : test
  user      : tester
  password  : testing

start rgw on http://localhost:8000
/home/danielbevenius/work/ceph/ceph/build/bin/radosgw -c /home/danielbevenius/work/ceph/ceph/build/ceph.conf --log-file=/home/danielbevenius/work/ceph/ceph/build/out/radosgw.8000.log --admin-socket=/home/danielbevenius/work/ceph/ceph/build/out/radosgw.8000.asok --pid-file=/home/danielbevenius/work/ceph/ceph/build/out/radosgw.8000.pid --debug-rgw=20 --debug-ms=1 -n client.rgw.8000 '--rgw_frontends=beast port=8000' 
vstart cluster complete. Use stop.sh to stop. See out/* (e.g. 'tail -f out/????') for debug output.

dashboard urls: https://127.0.0.1:41818
  w/ user/pass: admin / admin
restful urls: https://127.0.0.1:42818
  w/ user/pass: admin / b8937d0e-08f4-4ba0-a9f3-05ac3a4e93fe

Notice that you can copy the access key and secret key which can be use to connect to the S3 interface/API. We also see the url and port the rados gateway listens to (https://127.0.0.1:41818).

After this the status of the cluster can be checked using:

$ ./bin/ceph -s
 cluster:
    id:     bf1a5449-b6c8-4af8-9ee7-9edf7f2a8782
    health: HEALTH_WARN
            7 pool(s) have no replicas configured
 
  services:
    mon: 1 daemons, quorum a (age 72s)
    mgr: x(active, since 67s)
    mds: a:1 {0=a=up:active}
    osd: 1 osds: 1 up (since 42s), 1 in (since 42s)
    rgw: 1 daemon active (8000)
 
  task status:
    scrub status:
        mds.a: idle
 
  data:
    pools:   7 pools, 193 pgs
    objects: 223 objects, 9.5 KiB
    usage:   2.0 GiB used, 99 GiB / 101 GiB avail
    pgs:     193 active+clean
 
  progress:
    PG autoscaler decreasing pool 7 PGs from 32 to 8 (0s)
      [............................] 

$ ../src/stop.sh

$ sudo ./cephadm rm-cluster --fsid=62418068-9a65-11ea-b0db-482ae34923e7 --force

You can use ./cephadm ls to get the fsid.

$ sudo dnf install -y librados-devel

The n-api directory contains an example of what a node native addon written using N-API might look like.

This native module use Librados.

$ cd n-api
$ npm i
$ npm run compile
$ node index.js
Ceph init...
Ceph constructor...
librados version: 3.0.0
ceph-napi init...
Connected to Ceph cluster
Is io_ctx valid: true
Successfully wrote to pool data.
Successfully read data: bajja
init callback:  connected

This section should look at how a Ceph REST gateway is set up and create an example client that uses it. The Ceph Object Gateway Daemon (radosgw) is a http server which provides access to a Ceph storage cluster. The REST APIs provided are compatible with both Amazon S3 and OpenStack Swift.

We think that users will be most comfortable with Amazon S3 API so the example here will show how to use to configure the Amazon S3 API to work against radosgw.

A very basic example can be found in rest/index.js showing how one can create a bucket and poplate it.

$ cd rest
$ npm i
$ node index.js

The motivation for having a native node module for Ceph is that, from what I understand, a Ceph client is cluster aware and can communicate directly with OSDs to read and write data. This could perhaps be useful in situaions where a user wants to implement a function as a service that responds to some Knative event. The function in this use case would not be exposed to the outside and could use the native Ceph node module instead of using a REST based client.

A REST based client would be useful in other cases where user already familiar with the S3 API and Swift. For most of these cases users should be able to use the Amazon S3 SDK to interact with a Ceph cluster.