This commit expands the usage of the caching layer
for auth server API:
* Introduces in-memory cache that is used to serve all
Auth server API requests. This is done to achieve scalability
on 10K+ node clusters, where each node fetches certificate authorities,
roles, users and join tokens. It is not possible to scale
DynamoDB backend or other backends on 10K reads per seconds
on a single shard or partition. The solution is to introduce
an in-memory cache of the backend state that is always used
for reads.
* In-memory cache has been expanded to support all resources
required by the auth server.
* Experimental `tctl top` command has been introduced to display
common single node metrics.
Replace SQLite Memory Backend with BTree
SQLite in memory backend was suffering from
high tail latencies under load (up to 8 seconds
in 99.9%-ile on load configurations).
This commit replaces the SQLite memory caching
backend with in-memory BTree backend that
brought down tail latencies to 2 seconds (99.9%-ile)
and brought overall performance improvement.
Added utils.CertChecker that wraps a ssh.CertChecker. The new
certificate checker first checks if the certificate is a valid
certificate for Teleport. At the moment that is 2048-bit RSA then calls
the underlying certificate checker to perform the requested validation.
This commit introduces several key changes to
Teleport backend and API infrastructure
in order to achieve scalability improvements
on 10K+ node deployments.
Events and plain keyspace
--------------------------
New backend interface supports events,
pagination and range queries
and moves away from buckets to
plain keyspace, what better aligns
with DynamoDB and Etcd featuring similar
interfaces.
All backend implementations are
exposing Events API, allowing
multiple subscribers to consume the same
event stream and avoid polling database.
Replacing BoltDB, Dir with SQLite
-------------------------------
BoltDB backend does not support
having two processes access the database at the
same time. This prevented Teleport
using BoltDB backend to be live reloaded.
SQLite supports reads/writes by multiple
processes and makes Dir backend obsolete
as SQLite is more efficient on larger collections,
supports transactions and can detect data
corruption.
Teleport automatically migrates data from
Bolt and Dir backends into SQLite.
GRPC API and protobuf resources
-------------------------------
GRPC API has been introduced for
the auth server. The auth server now serves both GRPC
and JSON-HTTP API on the same TLS socket and uses
the same client certificate authentication.
All future API methods should use GRPC and HTTP-JSON
API is considered obsolete.
In addition to that some resources like
Server and CertificateAuthority are now
generated from protobuf service specifications in
a way that is fully backward compatible with
original JSON spec and schema, so the same resource
can be encoded and decoded from JSON, YAML
and protobuf.
All models should be refactored
into new proto specification over time.
Streaming presence service
--------------------------
In order to cut bandwidth, nodes
are sending full updates only when changes
to labels or spec have occured, otherwise
new light-weight GRPC keep alive updates are sent
over to the presence service, reducing
bandwidth usage on multi-node deployments.
In addition to that nodes are no longer polling
auth server for certificate authority rotation
updates, instead they subscribe to event updates
to detect updates as soon as they happen.
This is a new API, so the errors are inevitable,
that's why polling is still done, but
on a way slower rate.
This commit moves proxy kubernetes configuration
to a separate nested block to provide more fine
grained settings:
```yaml
auth:
kubernetes_ca_cert_path: /tmp/custom-ca
proxy:
enabled: yes
kubernetes:
enabled: yes
public_addr: [custom.example.com:port]
api_addr: kuberentes.example.com:443
listen_addr: localhost:3026
```
1. Kubernetes config section is explicitly enabled
and disabled. It is disabled by default.
2. Public address in kubernetes section
is propagated to tsh profile
The other part of the commit updates Ping
endpoint to send proxy configuration back to
the client, including kubernetes public address
and ssh listen address.
Clients updates profile accordingly to configuration
received from the proxy.
This commit implements #1935, fixes#2038
Auth server now supports global
defaults for timeout behavior:
```
auth_service:
client_idle_timeout: 15m
disconnect_expired_cert: no
```
New role options were introduced:
```
kind: role
version: v3
metadata:
name: intern
spec:
options:
# these two settings override the global ones:
client_idle_timeout: 1m
disconnect_expired_cert: yes
```
The following changes have been introduced
to tsh login behavior:
1. tsh login now accepts cluster name
as an optional positional argument:
$ tsh login clustername
2. If tsh login is called without arguments
and the current credentials are valid,
tsh login now prints status, previous behavior
always forced login:
$ tsh login
... print status if logged in...
2. If tsh login is called with the proxy
equal to current, tsh login selects cluster,
otherwise it will re-login to another proxy:
$ tsh login one
... selected cluster one
$ tsh login two
... selected cluster two
$ tsh login --proxy=example.com three
... selected cluster three because
proxy is the same
$ tsh login --proxy=acme.example.com four
...will switch to proxy acme.example.com
and cluster four
This commit implements #1860
During the the rotation procedure issuing TLS and SSH
certificate authorities are re-generated and all internal
components of the cluster re-register to get new
credentials.
The rotation procedure is based on a distributed
state machine algorithm - certificate authorities have
explicit rotation state and all parts of the cluster sync
local state machines by following transitions between phases.
Operator can launch CA rotation in auto or manual modes.
In manual mode operator moves cluster bewtween rotation states
and watches the states of the components to sync.
In auto mode state transitions are happening automatically
on a specified schedule.
The design documentation is embedded in the code:
lib/auth/rotate.go
Updates #1755
Design
------
This commit adds support for pluggable events and
sessions recordings and adds several plugins.
In case if external sessions recording storage
is used, nodes or proxies depending on configuration
store the session recordings locally and
then upload the recordings in the background.
Non-print session events are always sent to the
remote auth server as usual.
In case if remote events storage is used, auth
servers download recordings from it during playbacks.
DynamoDB event backend
----------------------
Transient DynamoDB backend is added for events
storage. Events are stored with default TTL of 1 year.
External lambda functions should be used
to forward events from DynamoDB.
Parameter audit_table_name in storage section
turns on dynamodb backend.
The table will be auto created.
S3 sessions backend
-------------------
If audit_sessions_uri is specified to s3://bucket-name
node or proxy depending on recording mode
will start uploading the recorded sessions
to the bucket.
If the bucket does not exist, teleport will
attempt to create a bucket with versioning and encryption
turned on by default.
Teleport will turn on bucket-side encryption for the tarballs
using aws:kms key.
File sessions backend
---------------------
If audit_sessions_uri is specified to file:///folder
teleport will start writing tarballs to this folder instead
of sending records to the file server.
This is helpful for plugin writers who can use fuse or NFS
mounted storage to handle the data.
Working dynamic configuration.
This commit introduces signal handling.
Parent teleport process is now capable of forking
the child process and passing listeners file descriptors
to the child.
Parent process then can gracefully shutdown
by tracking the amount of current connections and
closing listeners once the amount goes to 0.
Here are the signals handled:
* USR2 signal will cause the parent to fork
a child process and pass listener file descriptors to it.
Child process will close unused file descriptors
and will bind to the used ones.
At this moment two processes - the parent
and the forked child process will be serving requests.
After looking at the traffic and the log files,
administrator can either shut down the parent process
or the child process if the child process is not functioning
as expected.
* TERM, INT signals will trigger graceful process shutdown.
Auth, node and proxy processes will wait until the amount
of active connections goes down to 0 and will exit after that.
* KILL, QUIT signals will cause immediate non-graceful
shutdown.
* HUP signal combines USR2 and TERM signals in a convenient
way: parent process will fork a child process and
self-initate graceful shutdown. This is a more convenient
than USR2/TERM sequence, but less agile and robust
as if the connection to the parent process drops, but
the new process exits with error, administrators
can lock themselves out of the environment.
Additionally, boltdb backend has to be phased out,
as it does not support read/writes by two concurrent
processes. This had required refactoring of the dir
backend to use file locking to allow inter-process
collaboration on read/write operations.
Support configuration for web and reverse tunnel
proxies to listen on the same port.
* Default config are not changed for backwards compatibility.
* If administrator configures web and reverse tunnel
addresses to be on the same port, multiplexing is turned on
* In trusted clusters configuration reverse_tunnel_addr
defaults to web_addr.
This commit introduced mutual TLS authentication
for auth server API server.
Auth server multiplexes HTTP over SSH - existing
protocol and HTTP over TLS - new protocol
on the same listening socket.
Nodes and users authenticate with 2.5.0 Teleport
using TLS mutual TLS except backwards-compatibility
cases.
Instead of quietly changing behavior because `DEBUG` envar was set to
true, Teleport now explicitly requires scary --insecure flag to enable
this behavior.