Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

IP/Port configuration Proposal #143

Merged
merged 8 commits into from
May 2, 2023
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
203 changes: 203 additions & 0 deletions api-http/proposals/http-plugin.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,203 @@
# Proposal for Nodeos Endpoint IP and Port Configuration
This is a proposal to allow setting the IP and port for each endpoint provided by nodeos independently, i.e. address [IS #965](https://github.com/AntelopeIO/leap/issues/965).
## API category classification
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How do users discover what categories exist? This is especially important due to the behavior being "opt out": I need to know when I enable some plugin the exhaustive list categories I need to go in and potentially disable (it's possible this even changes on new releases, meaning in 4.0 I configured the entire set of categories for a plugin, but then in 5.0 a new category is introduced that becomes default-enabled)

I like how some of the categories are named after the plugin and then ro/rw: chain_ro and chain_rw, or net_ro and net_rw. These are "predictable" in that their name is shared with the plugin and there is a pretty clear delineation between danger zones. Something like producer_ro, producer_rw, and snapshot is more problematic since all three of those are from the same plugin.


Although [IS #965] (https://github.com/AntelopeIO/leap/issues/965) mentioned the *endpoint* as the unit of address configuration, if we allow each endpoint to be configurable, it
would be too fine-grained and verbose to use. Yet we feel that granularity at the plugin level might not meet user expectations. Therefore, we introduce a category level as the unit of server address configuration. The following are our proposed category classifications.

### chain_ro
- v1/chain/get_activated_protocol_features
- v1/chain/get_block_info
- v1/chain/get_block_header_state
- v1/chain/get_account
- v1/chain/get_code
- v1/chain/get_code_hash
- v1/chain/get_consensus_parameters
- v1/chain/get_abi
- v1/chain/get_raw_code_and_abi
- v1/chain/get_raw_abi
- v1/chain/get_table_rows
- v1/chain/get_table_by_scope
- v1/chain/get_currency_balance
- v1/chain/get_currency_stats
- v1/chain/get_producers
- v1/chain/get_producer_schedule
- v1/chain/get_scheduled_transactions
- v1/chain/get_required_keys
- v1/chain/get_transaction_id
- v1/chain/send_read_only_transaction
- v1/chain/get_accounts_by_authorizers
- v1/chain/get_raw_block
- v1/chain/get_block
- v1/chain/get_block_header
- v1/chain/get_transaction_status

### chain_rw
- v1/chain/compute_transaction
- v1/chain/push_transaction
- v1/chain/push_transactions
- v1/chain/send_transaction
- v1/chain/send_transaction2

### db_size
- v1/db_size/get
### net_ro
- v1/net/status
- v1/net/connections

### net_rw
- v1/net/connect
- v1/net/disconnect

### producer_ro
- v1/producer/paused
- v1/producer/get_runtime_options
- v1/producer/get_greylist
- v1/producer/get_whitelist_blacklist
- v1/producer/get_scheduled_protocol_feature_activations
- v1/producer/get_supported_protocol_features
- v1/producer/get_account_ram_corrections
- v1/producer/get_unapplied_transactions

### producer_rw
- v1/producer/pause
- v1/producer/resume
- v1/producer/update_runtime_options
- v1/producer/add_greylist_accounts
- v1/producer/remove_greylist_accounts
- v1/producer/set_whitelist_blacklist
- v1/producer/schedule_protocol_feature_activations
- v1/producer/get_integrity_hash

### snapshot
- v1/producer/get_snapshot_requests
- v1/producer/create_snapshot
- v1/producer/schedule_snapshot
- v1/producer/unschedule_snapshot

### trace_api
- v1/trace_api/get_block
- v1/trace_api/get_transaction_trace

### prometheus
- v1/prometheus/metrics

### node
- v1/node/get_supported_apis
- v1/chain/get_info

## User configuration knob
A single `http-category-address` can be used to configure all addresses in command line and ini file. The option can be used multiple times as needed.

```config.ini
http-server-address = http-category-address
http-category-address = chain_ro,127.0.0.1:8081
http-category-address = chain_ro,localhost:9090
http-category-address = chain_rw,[::1]:8082
http-category-address = net_ro,127.0.0.1:8081
http-category-address = net_rw,[::]:8083
http-category-address = producer_ro,/tmp/absolute_unix_path.sock
http-category-address = producer_rw,./relative/unix_path.sock
http-category-address = snapshot,./relative/unix_path.sock
```

A single `hostname:port` specification can be used by multiple categories, as of the case for `127.0.0.1:8081` in above example. However, two specifications having the same port with different hostname strings are always considered as configuration error regardless whether they can be resolved into the same set of IP addresses. For example, the following configuration are invalid.

```
http-category-address = chain_ro,127.0.0.1:8081
http-category-address = chain_rw,localhost:8081 # error, 127.0.0.1 and localhost are different
```

For the node category, it is NOT user configurable because it will be available for all listened http addresses.

For backward compatibility, the HTTP category facility has to be opted in by specifying `http-server-address = http-category-address` explicitly.
When `http-server-address = http-category-address`, all the unspecified categories are considered disabled. In addition, `http-server-address = http-category-address`
connot be specified together with a non-empty `unix-socket-path`.


All existing configuration options in the http plugin apply to all handlers for all addresses; this includes the `http-threads` option. This means all handlers share the same thread pool.

## API registration from other plugins

Existing API registration method in http_plugin
```
using url_path_t = std::string;
using request_body_t = std::string;
using url_handler = std::function<void(url_path_t&&, request_body_t&&, url_response_callback&&)>;
using api_entry = std::pair<url_path_t, url_handler>;
class http_plugin : public appbase::plugin<http_plugin>
{
public:

void add_handler(const string& url, const url_handler&, appbase::exec_queue q, int priority = appbase::priority::medium_low, http_content_type content_type = http_content_type::json);
void add_api(const api_description& api, appbase::exec_queue q, int priority = appbase::priority::medium_low, http_content_type content_type = http_content_type::json) {
for (const auto& call : api)
add_handler(call.first, call.second, q, priority, content_type);
}

void add_async_handler(const string& url, const url_handler& handler, http_content_type content_type = http_content_type::json);
void add_async_api(const api_description& api, http_content_type content_type = http_content_type::json) {
for (const auto& call : api)
add_async_handler(call.first, call.second, content_type);
}
}
```

It will be converted as follows in this proposal
```
using server_address_t = std::string;
using url_path_t = std::string;
using request_body_t = std::string;
using url_handler = std::function<void(const server_address_t&, url_path_t&&, request_body_t&&, url_response_callback&&)>;

enum class api_category {
node,
get_info,
chain_ro,
chain_rw,
net_ro,
net_rw,
producer_ro,
producer_rw,
snapshot,
trace_api,
prometheus,
};

struct api_entry {
api_category category;
std::string server_address; // empty value mean the API is available for all listened addresses, NOT DISABLE
std::string path;
url_handler handler;
appbase::exec_queue q;
int priority = appbase::priority::medium_low
http_content_type content_type = http_content_type::json;
};

class http_plugin : public appbase::plugin<http_plugin> {
public:
void add_sync_api(api_entry&& api);
void add_async_api(api_entry&& api);
};
```

## server address format
* server address can be in one of the following formats
- ipv4_address:port like 127.0.0.1:8080
- ipv6_address:port like [2001:db8:3c4d:15::1a2f:1a2b]:8080
- hostname:port like my.domain.com:8080
- unix socket path (must starts with '/' or './')

* In the case of the hostname, the IP address is resolved by ARP, *ALL* resolved addresses will be listened to.
spoonincode marked this conversation as resolved.
Show resolved Hide resolved
* To listen to all interfaces, just use `0.0.0.0:8080`, `[::]:8080` or `:8080`.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We talked about this some in the review and it wasn't entirely clear to me where we landed. I take it from this, that

  • 0.0.0.0:8080 -- listen on all IPv4 addresses
  • [::]:8080 -- listen on all IPv6 addresses
  • :8080 -- listen on all IPv4 & IPv6 addresses

In particular it sounded like we would not enable 4-in-6 behavior? In other words, we will expressly disable 4-in-6.


### How to handle IPv4-mapped IPv6?
There are a couple of options:
- Always use system default. That is, on Linux where IPv4-mapped IPv6 is enabled by default, `[::]:8080` will listen to both ipv4 and ipv6.
- Always disable IPv4-mapped IPv6. That is, `[::]:8080` will only listen to ipv6; `:8080` will listen on both ipv4 and ipv6.
- Use system default and provide option to be more specific, like `[::],ip-v6-only=on` or `[::],ip-v6-only=off`.

Apache server adopts the first option and provides a global "--enable-v4-mapped" option such that the behavior on platform where IPv4-mapped IPv6 is disabled by default can match others.

Ngnix, on the other hand, provides `ip-v6-only=on` and `ip-v6-only=off` to be specific, yet it uses `ip-v6-only=off` by default regardless of the platform it runs on since version 1.3.4.