This is like a Möbius strip of code due to the fact that low-level components (serf/memberlist) are connected to high-level components (the catalog and mesh-gateways) in a twisty maze of references which make it hard to dive into. With that in mind here's a high level summary of what you'll find in the patch:
There are several distinct chunks of code that are affected:
* new flags and config options for the server
* retry join WAN is slightly different
* retry join code is shared to discover primary mesh gateways from secondary datacenters
* because retry join logic runs in the *agent* and the results of that
operation for primary mesh gateways are needed in the *server* there are
some methods like `RefreshPrimaryGatewayFallbackAddresses` that must occur
at multiple layers of abstraction just to pass the data down to the right
layer.
* new cache type `FederationStateListMeshGatewaysName` for use in `proxycfg/xds` layers
* the function signature for RPC dialing picked up a new required field (the
node name of the destination)
* several new RPCs for manipulating a FederationState object:
`FederationState:{Apply,Get,List,ListMeshGateways}`
* 3 read-only internal APIs for debugging use to invoke those RPCs from curl
* raft and fsm changes to persist these FederationStates
* replication for FederationStates as they are canonically stored in the
Primary and replicated to the Secondaries.
* a special derivative of anti-entropy that runs in secondaries to snapshot
their local mesh gateway `CheckServiceNodes` and sync them into their upstream
FederationState in the primary (this works in conjunction with the
replication to distribute addresses for all mesh gateways in all DCs to all
other DCs)
* a "gateway locator" convenience object to make use of this data to choose
the addresses of gateways to use for any given RPC or gossip operation to a
remote DC. This gets data from the "retry join" logic in the agent and also
directly calls into the FSM.
* RPC (`:8300`) on the server sniffs the first byte of a new connection to
determine if it's actually doing native TLS. If so it checks the ALPN header
for protocol determination (just like how the existing system uses the
type-byte marker).
* 2 new kinds of protocols are exclusively decoded via this native TLS
mechanism: one for ferrying "packet" operations (udp-like) from the gossip
layer and one for "stream" operations (tcp-like). The packet operations
re-use sockets (using length-prefixing) to cut down on TLS re-negotiation
overhead.
* the server instances specially wrap the `memberlist.NetTransport` when running
with gateway federation enabled (in a `wanfed.Transport`). The general gist is
that if it tries to dial a node in the SAME datacenter (deduced by looking
at the suffix of the node name) there is no change. If dialing a DIFFERENT
datacenter it is wrapped up in a TLS+ALPN blob and sent through some mesh
gateways to eventually end up in a server's :8300 port.
* a new flag when launching a mesh gateway via `consul connect envoy` to
indicate that the servers are to be exposed. This sets a special service
meta when registering the gateway into the catalog.
* `proxycfg/xds` notice this metadata blob to activate additional watches for
the FederationState objects as well as the location of all of the consul
servers in that datacenter.
* `xds:` if the extra metadata is in place additional clusters are defined in a
DC to bulk sink all traffic to another DC's gateways. For the current
datacenter we listen on a wildcard name (`server.<dc>.consul`) that load
balances all servers as well as one mini-cluster per node
(`<node>.server.<dc>.consul`)
* the `consul tls cert create` command got a new flag (`-node`) to help create
an additional SAN in certs that can be used with this flavor of federation.
Previously this happened to be validating only the chains in the default namespace. Now it will validate all chains in all namespaces when the global proxy-defaults is changed.
The previous value was too conservative and users with many instances
were having problems because of it. This change increases the limit to
8192 which reportedly fixed most of the issues with that.
Related: #4984, #4986, #5050.
* Renamed structs.IntentionWildcard to structs.WildcardSpecifier
* Refactor ACL Config
Get rid of remnants of enterprise only renaming.
Add a WildcardName field for specifying what string should be used to indicate a wildcard.
* Add wildcard support in the ACL package
For read operations they can call anyAllowed to determine if any read access to the given resource would be granted.
For write operations they can call allAllowed to ensure that write access is granted to everything.
* Make v1/agent/connect/authorize namespace aware
* Update intention ACL enforcement
This also changes how intention:read is granted. Before the Intention.List RPC would allow viewing an intention if the token had intention:read on the destination. However Intention.Match allowed viewing if access was allowed for either the source or dest side. Now Intention.List and Intention.Get fall in line with Intention.Matches previous behavior.
Due to this being done a few different places ACL enforcement for a singular intention is now done with the CanRead and CanWrite methods on the intention itself.
* Refactor Intention.Apply to make things easier to follow.
Restore a few more service-kind index updates so blocking in ServiceDump works in more cases
Namely one omission was that check updates for dumped services were not
unblocking.
Also adds a ServiceDump state store test and also fix a watch bug with the
normal dump.
Follow-on from #6916
Ensure we close the Sentinel Evaluator so as not to leak go routines
Fix a bunch of test logging so that various warnings when starting a test agent go to the ltest logger and not straight to stdout.
Various canned ent meta types always return a valid pointer (no more nils). This allows us to blindly deref + assign in various places.
Update ACL index tracking to ensure oss -> ent upgrades will work as expected.
Update ent meta parsing to include function to disallow wildcarding.
Main Changes:
• method signature updates everywhere to account for passing around enterprise meta.
• populate the EnterpriseAuthorizerContext for all ACL related authorizations.
• ACL resource listings now operate like the catalog or kv listings in that the returned entries are filtered down to what the token is allowed to see. With Namespaces its no longer all or nothing.
• Modified the acl.Policy parsing to abstract away basic decoding so that enterprise can do it slightly differently. Also updated method signatures so that when parsing a policy it can take extra ent metadata to use during rules validation and policy creation.
Secondary Changes:
• Moved protobuf encoding functions out of the agentpb package to eliminate circular dependencies.
• Added custom JSON unmarshalers for a few ACL resource types (to support snake case and to get rid of mapstructure)
• AuthMethod validator cache is now an interface as these will be cached per-namespace for Consul Enterprise.
• Added checks for policy/role link existence at the RPC API so we don’t push the request through raft to have it fail internally.
• Forward ACL token delete request to the primary datacenter when the secondary DC doesn’t have the token.
• Added a bunch of ACL test helpers for inserting ACL resource test data.
* ACL Authorizer overhaul
To account for upcoming features every Authorization function can now take an extra *acl.EnterpriseAuthorizerContext. These are unused in OSS and will always be nil.
Additionally the acl package has received some thorough refactoring to enable all of the extra Consul Enterprise specific authorizations including moving sentinel enforcement into the stubbed structs. The Authorizer funcs now return an acl.EnforcementDecision instead of a boolean. This improves the overall interface as it makes multiple Authorizers easily chainable as they now indicate whether they had an authoritative decision or should use some other defaults. A ChainedAuthorizer was added to handle this Authorizer enforcement chain and will never itself return a non-authoritative decision.
* Include stub for extra enterprise rules in the global management policy
* Allow for an upgrade of the global-management policy
* Add build system support for protobuf generation
This is done generically so that we don’t have to keep updating the makefile to add another proto generation.
Note: anything not in the vendor directory and with a .proto extension will be run through protoc if the corresponding namespace.pb.go file is not up to date.
If you want to rebuild just a single proto file you can do so with: make proto-rebuild PROTOFILES=<list of proto files to rebuild>
Providing the PROTOFILES var will override the default behavior of finding all the .proto files.
* Start adding types to the agent/proto package
These will be needed for some other work and are by no means comprehensive.
* Add ability to resolve/fixup the agentpb.ACLLinks structure in the state store.
* Use protobuf marshalling of raft requests instead of msgpack for protoc generated types.
This does not change any encoding of existing types.
* Removed structs package automatically encoding with protobuf marshalling
Instead the caller of raftApply that wants to opt-in to protobuf encoding will have to call `raftApplyProtobuf`
* Run update-vendor to fixup modules.txt
Nothing changed as far as dependencies go but the ordering of modules in that file depends on the time they are first seen and its not alphabetical.
* Rename some things and implement the structs.RPCInfo interface bits
agentpb.QueryOptions and agentpb.WriteRequest implement 3 of the 4 RPCInfo funcs and the new TargetDatacenter message type implements the fourth.
* Use the right encoding function.
* Renamed agent/proto package to agent/agentpb to prevent package name conflicts
* Update modules.txt to fix ordering
* Change blockingQuery to take in interfaces for the query options and meta
* Add %T to error output.
* Add/Update some comments
When there is an node name conflicts, such messages are displayed within Consul:
`consul.fsm: EnsureRegistration failed: failed inserting node: Error while renaming Node ID: "e1d456bc-f72d-98e5-ebb3-26ae80d785cf": Node name node001 is reserved by node 05f10209-1b9c-b90c-e3e2-059e64556d4a with name node001`
While it is easy to find the node that has reserved the name, it is hard to find
the node trying to aquire the name since it is not registered, because it
is not part of `consul members` output
This PR will display the IP of the offender and solve far more easily those issues.
Failover is pushed entirely down to the data plane by creating envoy
clusters and putting each successive destination in a different load
assignment priority band. For example this shows that normally requests
go to 1.2.3.4:8080 but when that fails they go to 6.7.8.9:8080:
- name: foo
load_assignment:
cluster_name: foo
policy:
overprovisioning_factor: 100000
endpoints:
- priority: 0
lb_endpoints:
- endpoint:
address:
socket_address:
address: 1.2.3.4
port_value: 8080
- priority: 1
lb_endpoints:
- endpoint:
address:
socket_address:
address: 6.7.8.9
port_value: 8080
Mesh gateways route requests based solely on the SNI header tacked onto
the TLS layer. Envoy currently only lets you configure the outbound SNI
header at the cluster layer.
If you try to failover through a mesh gateway you ideally would
configure the SNI value per endpoint, but that's not possible in envoy
today.
This PR introduces a simpler way around the problem for now:
1. We identify any target of failover that will use mesh gateway mode local or
remote and then further isolate any resolver node in the compiled discovery
chain that has a failover destination set to one of those targets.
2. For each of these resolvers we will perform a small measurement of
comparative healths of the endpoints that come back from the health API for the
set of primary target and serial failover targets. We walk the list of targets
in order and if any endpoint is healthy we return that target, otherwise we
move on to the next target.
3. The CDS and EDS endpoints both perform the measurements in (2) for the
affected resolver nodes.
4. For CDS this measurement selects which TLS SNI field to use for the cluster
(note the cluster is always going to be named for the primary target)
5. For EDS this measurement selects which set of endpoints will populate the
cluster. Priority tiered failover is ignored.
One of the big downsides to this approach to failover is that the failover
detection and correction is going to be controlled by consul rather than
deferring that entirely to the data plane as with the prior version. This also
means that we are bound to only failover using official health signals and
cannot make use of data plane signals like outlier detection to affect
failover.
In this specific scenario the lack of data plane signals is ok because the
effectiveness is already muted by the fact that the ultimate destination
endpoints will have their data plane signals scrambled when they pass through
the mesh gateway wrapper anyway so we're not losing much.
Another related fix is that we now use the endpoint health from the
underlying service, not the health of the gateway (regardless of
failover mode).
All these changes should have no side-effects or change behavior:
- Use bytes.Buffer's String() instead of a conversion
- Use time.Since and time.Until where fitting
- Drop unnecessary returns and assignment
* Ensure the mesh gateway configuration comes back in the api within each upstream
* Add a test for the MeshGatewayConfig in the ToAPI functions
* Ensure we don’t use gateways for dc local connections
* Update the svc kind index for deletions
* Replace the proxycfg.state cache with an interface for testing
Also start implementing proxycfg state testing.
* Update the state tests to verify some gateway watches for upstream-targets of a discovery chain.
maxIndexWatchTxn was only watching the IndexEntry of the max index of all the entries. It needed to watch all of them regardless of which was the max.
Also plumbed the query source through in the proxy config to help better track requests.
The general problem was that a the CA config which contained the trust domain was happening outside of the blocking mechanism so if the client started the blocking query before the primary dcs roots had been set then a state trust domain was being pushed down.
This was fixed here but in the future we should probably fixup the CA initialization code to not initialize the CA config twice when it doesn’t need to.
With this you should be able to fetch all of the relevant discovery
chain config entries from the state store in one query and then feed
them into the compiler outside of a transaction.
There are a lot of TODOs scattered through here, but they're mostly
around handling fun edge cases and can be deferred until more of the
plumbing works completely.
If a KVSet is performed but does not update the entry, do not trigger
watches for this key.
This avoids releasing blocking queries for KV values that did not
actually changed.
Roles are named and can express the same bundle of permissions that can
currently be assigned to a Token (lists of Policies and Service
Identities). The difference with a Role is that it not itself a bearer
token, but just another entity that can be tied to a Token.
This lets an operator potentially curate a set of smaller reusable
Policies and compose them together into reusable Roles, rather than
always exploding that same list of Policies on any Token that needs
similar permissions.
This also refactors the acl replication code to be semi-generic to avoid
3x copypasta.
This is mainly to avoid having the API return "0001-01-01T00:00:00Z" as
a value for the ExpirationTime field when it is not set. Unfortunately
time.Time doesn't respect the json marshalling "omitempty" directive.
These act like a special cased version of a Policy Template for granting
a token the privileges necessary to register a service and its connect
proxy, and read upstreams from the catalog.
Fixes: #4222
# Data Filtering
This PR will implement filtering for the following endpoints:
## Supported HTTP Endpoints
- `/agent/checks`
- `/agent/services`
- `/catalog/nodes`
- `/catalog/service/:service`
- `/catalog/connect/:service`
- `/catalog/node/:node`
- `/health/node/:node`
- `/health/checks/:service`
- `/health/service/:service`
- `/health/connect/:service`
- `/health/state/:state`
- `/internal/ui/nodes`
- `/internal/ui/services`
More can be added going forward and any endpoint which is used to list some data is a good candidate.
## Usage
When using the HTTP API a `filter` query parameter can be used to pass a filter expression to Consul. Filter Expressions take the general form of:
```
<selector> == <value>
<selector> != <value>
<value> in <selector>
<value> not in <selector>
<selector> contains <value>
<selector> not contains <value>
<selector> is empty
<selector> is not empty
not <other expression>
<expression 1> and <expression 2>
<expression 1> or <expression 2>
```
Normal boolean logic and precedence is supported. All of the actual filtering and evaluation logic is coming from the [go-bexpr](https://github.com/hashicorp/go-bexpr) library
## Other changes
Adding the `Internal.ServiceDump` RPC endpoint. This will allow the UI to filter services better.
* Make Connect health queryies unblock correctly in all cases and use optimal number of watch chans. Fixes#5506.
* Node check test cases and clearer bug test doc
* Comment update
Refs #4984.
Watching chans for every node we touch in a health query is wasteful. In #4984 it shows that if there are more than 682 service instances we always fallback to watching all services which kills performance.
We already have a record in MemDB that is reliably update whenever the service health result should change thanks to per-service watch indexes.
So in general, provided there is at least one service instances and we actually have a service index for it (we always do now) we only ever need to watch a single channel.
This saves us from ever falling back to the general index and causing the performance cliff in #4984, but it also means fewer goroutines and work done for every blocking health query.
It also saves some allocations made during the query because we no longer have to populate a WatchSet with 3 chans per service instance which saves the internal map allocation.
This passes all state store tests except the one that explicitly checked for the fallback behaviour we've now optimized away and in general seems safe.
Node updates were not updating the service indexes, which are used for
service related queries. This caused the X-Consul-Index to stay the same
after a node update as seen from a service query even though the node
data is returned in heath queries. If that happened in between queries
the client would miss this change.
We now update the indexes of the services on the node when it is
updated.
Fixes: #5450
Previously we were fixing up the token links directly on the *ACLToken returned by memdb. This invalidated some assumptions that a snapshot is immutable as well as potentially being able to cause a crash.
The fix here is to give the policy link fixing function copy on write semantics. When no fixes are necessary we can return the memdb object directly, otherwise we copy it and create a new list of links.
Eventually we might find a better way to keep those policy links in sync but for now this fixes the issue.
* Support rate limiting and concurrency limiting CSR requests on servers; handle CA rotations gracefully with jitter and backoff-on-rate-limit in client
* Add CSR rate limiting docs
* Fix config naming and add tests for new CA configs
* Store leaf cert indexes in raft and use for the ModifyIndex on the returned certs
This ensures that future certificate signings will have a strictly greater ModifyIndex than any previous certs signed.
## Background
When making a blocking query on a missing service (was never registered, or is not registered anymore) the query returns as soon as any service is updated.
On clusters with frequent updates (5~10 updates/s in our DCs) these queries virtually do not block, and clients with no protections againt this waste ressources on the agent and server side. Clients that do protect against this get updates later than they should because of the backoff time they implement between requests.
## Implementation
While reducing the number of unnecessary updates we still want :
* Clients to be notified as soon as when the last instance of a service disapears.
* Clients to be notified whenever there's there is an update for the service.
* Clients to be notified as soon as the first instance of the requested service is added.
To reduce the number of unnecessary updates we need to block when a request to a missing service is made. However in the following case :
1. Client `client1` makes a query for service `foo`, gets back a node and X-Consul-Index 42
2. `foo` is unregistered
3. `client1` makes a query for `foo` with `index=42` -> `foo` does not exist, the query blocks and `client1` is not notified of the change on `foo`
We could store the last raft index when each service was last alive to know wether we should block on the incoming query or not, but that list could grow indefinetly.
We instead store the last raft index when a service was unregistered and use it when a query targets a service that does not exist.
When a service `srv` is unregistered this "missing service index" is always greater than any X-Consul-Index held by the clients while `srv` was up, allowing us to immediatly notify them.
1. Client `client1` makes a query for service `foo`, gets back a node and `X-Consul-Index: 42`
2. `foo` is unregistered, we set the "missing service index" to 43
3. `client1` makes a blocking query for `foo` with `index=42` -> `foo` does not exist, we check against the "missing service index" and return immediatly with `X-Consul-Index: 43`
4. `client1` makes a blocking query for `foo` with `index=43` -> we block
5. Other changes happen in the cluster, but foo still doesn't exist and "missing service index" hasn't changed, the query is still blocked
6. `foo` is registered again on index 62 -> `foo` exists and its index is greater than 43, we unblock the query
This PR both prevents a blank CA config from being written out to
a snapshot and allows Consul to gracefully recover from a snapshot
with an invalid CA config.
Fixes#4954.
* Add leader token upgrade test and fix various ACL enablement bugs
* Update the leader ACL initialization tests.
* Add a StateStore ACL tests for ACLTokenSet and ACLTokenGetBy* functions
* Advertise the agents acl support status with the agent/self endpoint.
* Make batch token upsert CAS’able to prevent consistency issues with token auto-upgrade
* Finish up the ACL state store token tests
* Finish the ACL state store unit tests
Also rename some things to make them more consistent.
* Do as much ACL replication testing as I can.
This PR is almost a complete rewrite of the ACL system within Consul. It brings the features more in line with other HashiCorp products. Obviously there is quite a bit left to do here but most of it is related docs, testing and finishing the last few commands in the CLI. I will update the PR description and check off the todos as I finish them over the next few days/week.
Description
At a high level this PR is mainly to split ACL tokens from Policies and to split the concepts of Authorization from Identities. A lot of this PR is mostly just to support CRUD operations on ACLTokens and ACLPolicies. These in and of themselves are not particularly interesting. The bigger conceptual changes are in how tokens get resolved, how backwards compatibility is handled and the separation of policy from identity which could lead the way to allowing for alternative identity providers.
On the surface and with a new cluster the ACL system will look very similar to that of Nomads. Both have tokens and policies. Both have local tokens. The ACL management APIs for both are very similar. I even ripped off Nomad's ACL bootstrap resetting procedure. There are a few key differences though.
Nomad requires token and policy replication where Consul only requires policy replication with token replication being opt-in. In Consul local tokens only work with token replication being enabled though.
All policies in Nomad are globally applicable. In Consul all policies are stored and replicated globally but can be scoped to a subset of the datacenters. This allows for more granular access management.
Unlike Nomad, Consul has legacy baggage in the form of the original ACL system. The ramifications of this are:
A server running the new system must still support other clients using the legacy system.
A client running the new system must be able to use the legacy RPCs when the servers in its datacenter are running the legacy system.
The primary ACL DC's servers running in legacy mode needs to be a gate that keeps everything else in the entire multi-DC cluster running in legacy mode.
So not only does this PR implement the new ACL system but has a legacy mode built in for when the cluster isn't ready for new ACLs. Also detecting that new ACLs can be used is automatic and requires no configuration on the part of administrators. This process is detailed more in the "Transitioning from Legacy to New ACL Mode" section below.
* Support multiple tags for health and catalog api endpoints
Fixes#1781.
Adds a `ServiceTags` field to the ServiceSpecificRequest to support
multiple tags, updates the filter logic in the catalog store, and
propagates these change through to the health and catalog endpoints.
Note: Leaves `ServiceTag` in the struct, since it is being used as
part of the DNS lookup, which in turn uses the health check.
* Update the api package to support multiple tags
Includes additional tests.
* Update new tests to use the `require` library
* Update HealthConnect check after a bad merge
* [Performance On Large clusters] Checks do update services/nodes only when really modified to avoid too many updates on very large clusters
In a large cluster, when having a few thousands of nodes, the anti-entropy
mechanism performs lots of changes (several per seconds) while
there is no real change. This patch wants to improve this in order
to increase Consul scalability when using many blocking requests on
health for instance.
* [Performance for large clusters] Only updates index of service if service is really modified
* [Performance for large clusters] Only updates index of nodes if node is really modified
* Added comments / ensure IsSame() has clear semantics
* Avoid having modified boolean, return nil directly if stutures are Same
* Fixed unstable unit tests TestLeader_ChangeServerID
* Rewrite TestNode_IsSame() for better readability as suggested by @banks
* Rename ServiceNode.IsSame() into IsSameService() + added unit tests
* Do not duplicate TestStructs_ServiceNode_Conversions() and increase test coverage of IsSameService
* Clearer documentation in IsSameService
* Take into account ServiceProxy into ServiceNode.IsSameService()
* Fixed IsSameService() with all new structures
* Refactor Service Definition ProxyDestination.
This includes:
- Refactoring all internal structs used
- Updated tests for both deprecated and new input for:
- Agent Services endpoint response
- Agent Service endpoint response
- Agent Register endpoint
- Unmanaged deprecated field
- Unmanaged new fields
- Managed deprecated upstreams
- Managed new
- Catalog Register
- Unmanaged deprecated field
- Unmanaged new fields
- Managed deprecated upstreams
- Managed new
- Catalog Services endpoint response
- Catalog Node endpoint response
- Catalog Service endpoint response
- Updated API tests for all of the above too (both deprecated and new forms of register)
TODO:
- config package changes for on-disk service definitions
- proxy config endpoint
- built-in proxy support for new fields
* Agent proxy config endpoint updated with upstreams
* Config file changes for upstreams.
* Add upstream opaque config and update all tests to ensure it works everywhere.
* Built in proxy working with new Upstreams config
* Command fixes and deprecations
* Fix key translation, upstream type defaults and a spate of other subtele bugs found with ned to end test scripts...
TODO: tests still failing on one case that needs a fix. I think it's key translation for upstreams nested in Managed proxy struct.
* Fix translated keys in API registration.
≈
* Fixes from docs
- omit some empty undocumented fields in API
- Bring back ServiceProxyDestination in Catalog responses to not break backwards compat - this was removed assuming it was only used internally.
* Documentation updates for Upstreams in service definition
* Fixes for tests broken by many refactors.
* Enable travis on f-connect branch in this branch too.
* Add consistent Deprecation comments to ProxyDestination uses
* Update version number on deprecation notices, and correct upstream datacenter field with explanation in docs
* Implementation of Weights Data structures
Adding this datastructure will allow us to resolve the
issues #1088 and #4198
This new structure defaults to values:
```
{ Passing: 1, Warning: 0 }
```
Which means, use weight of 0 for a Service in Warning State
while use Weight 1 for a Healthy Service.
Thus it remains compatible with previous Consul versions.
* Implemented weights for DNS SRV Records
* DNS properly support agents with weight support while server does not (backwards compatibility)
* Use Warning value of Weights of 1 by default
When using DNS interface with only_passing = false, all nodes
with non-Critical healthcheck used to have a weight value of 1.
While having weight.Warning = 0 as default value, this is probably
a bad idea as it breaks ascending compatibility.
Thus, we put a default value of 1 to be consistent with existing behaviour.
* Added documentation for new weight field in service description
* Better documentation about weights as suggested by @banks
* Return weight = 1 for unknown Check states as suggested by @banks
* Fixed typo (of -> or) in error message as requested by @mkeeler
* Fixed unstable unit test TestRetryJoin
* Fixed unstable tests
* Fixed wrong Fatalf format in `testrpc/wait.go`
* Added notes regarding DNS SRV lookup limitations regarding number of instances
* Documentation fixes and clarification regarding SRV records with weights as requested by @banks
* Rephrase docs
* Display more information about check being not properly added when it fails
It follows an incident where we add lots of error messages:
[WARN] consul.fsm: EnsureRegistration failed: failed inserting check: Missing service registration
That seems related to Consul failing to restart on respective agents.
Having Node information as well as service information would help diagnose the issue.
* Renamed ensureCheckIfNodeMatches() as requested by @banks
* Allow to rename nodes with IDs, will fix#3974 and #4413
This change allow to rename any well behaving recent agent with an
ID to be renamed safely, ie: without taking the name of another one
with case insensitive comparison.
Deprecated behaviour warning
----------------------------
Due to asceding compatibility, it is still possible however to
"take" the name of another name by not providing any ID.
Note that when not providing any ID, it is possible to have 2 nodes
having similar names with case differences, ie: myNode and mynode
which might lead to DB corruption on Consul server side and
lead to server not properly restarting.
See #3983 and #4399 for Context about this change.
Disabling registration of nodes without IDs as specified in #4414
should probably be the way to go eventually.
* Removed the case-insensitive search when adding a node within the else
block since it breaks the test TestAgentAntiEntropy_Services
While the else case is probably legit, it will be fixed with #4414 in
a later release.
* Added again the test in the else to avoid duplicated names, but
enforce this test only for nodes having IDs.
Thus most tests without any ID will work, and allows us fixing
* Added more tests regarding request with/without IDs.
`TestStateStore_EnsureNode` now test registration and renaming with IDs
`TestStateStore_EnsureNodeDeprecated` tests registration without IDs
and tests removing an ID from a node as well as updated a node
without its ID (deprecated behaviour kept for backwards compatibility)
* Do not allow renaming in case of conflict, including when other node has no ID
* Fixed function GetNodeID that was not working due to wrong type when searching node from its ID
Thus, all tests about renaming were not working properly.
Added the full test cas that allowed me to detect it.
* Better error messages, more tests when nodeID is not a valid UUID in GetNodeID()
* Added separate TestStateStore_GetNodeID to test GetNodeID.
More complete test coverage for GetNodeID
* Added new unit test `TestStateStore_ensureNoNodeWithSimilarNameTxn`
Also fixed comments to be clearer after remarks from @banks
* Fixed error message in unit test to match test case
* Use uuid.ParseUUID to parse Node.ID as requested by @mkeeler
Since DNS is case insensitive and DB as issues when similar names with different
cases are added, check for unicity based on case insensitivity.
Following another big incident we had in our cluster, we also validate
that adding/renaming a not does not conflicts with case insensitive
matches.
We had the following error once:
- one node called: mymachine.MYDC.mydomain was shut off
- another node (different ID) was added with name: mymachine.mydc.mydomain before
72 hours
When restarting the consul server of domain, the consul server restarted failed
to start since it detected an issue in RAFT database because
mymachine.MYDC.mydomain and mymachine.mydc.mydomain had the same names.
Checking at registration time with case insensitivity should definitly fix
those issues and avoid Consul DB corruption.
Intention de-duplication in previously merged PR actualy failed some tests that were not caught be me or CI. I ran the test files for state changes but they happened not to trigger this case so I made sure they did first and then fixed. That fixed some upstream intention endpoint tests that I'd not run as part of testing the previous fix.
Bugfix for https://github.com/hashicorp/consul/pull/3899
When a node level check is removed (example: maintenance),
some watchers on services might have to recompute their state.
If those nodes are performing blocking queries, they have to be notified.
While their state was updated when node-level state did change or was added
this was not the case when the check was removed. This fixes it.
This patch did give some better results, but break watches on
the services of a node.
It is possible to apply the same optimization for nodes than
to services (one index per instance), but it would complicate
further the patch.
Let's do it in another PR.
This patch improves the watches for services on large cluster:
each service has now its own index, such watches on a specific service
are not modified by changes in the global catalog.
It should improve a lot the performance of tools such as consul-template
or libraries performing watches on very large clusters with many
services/watches.
The lock isn't needed after we clean up the expire bin, and as seen
in #3700 we can get into a deadlock waiting to place the expire index
into the channel while holding this lock.
Fixes#3700
R.B. Boyer