HTTP Submission API
- Table of Contents
- Overview
- Example
- Compression
- Request Document
- Request for Multiple Messages
- HTTP keep-alive
- Message Acceptance Throttling and Back-Pressure
- Logging API Calls
Overview
GreenArrow Engine provides an HTTP message submission API.
The information is provided to the API as a POST or PUT request of JSON content. (The header of content-type: application/json must be provided.)
The API provides a response of JSON content with a content type of application/json.
The uploaded information can be compressed using gzip or deflate compression. The header of Content-Encoding: gzip or Content-Encoding: deflate must be provided when doing this.
There is a 10MB size limit per submission. This limit is applied before a compressed payload is uncompressed, so if, for example, a 25MB payload compresses down to 8MB, you can submit it, but if it compresses down to 11MB, you cannot.
Authentication is performed by providing a username/password for an Email User in the JSON document.
The API is provided at /api/v1/send.json on your GreenArrow server’s HTTP service.
In order to maximize your injection speed, we strongly recommend use of batching (see the Request for Multiple Messages section of this document) and HTTP keep-alive.
Example
Example JSON document:
{
"username":"[email protected]",
"password":"test",
"message":{
"html":"html content goes <b>here</b>",
"text":"text content goes here",
"subject":"this is the subject",
"to":[
{
"email":"[email protected]",
"name":"John Doe"
}
],
"from_email":"[email protected]",
"from_name":"Your Company"
}
}
Example JSON reply for success:
{"success":1,"message_id":"[email protected]"}
Example JSON reply for failure:
{"success":0,"error":"no data in POST or PUT payload"}
or:
{"success":0,"error":"incorrect username\/password"}
Submitting this request with curl:
cat <<'EOT' > post_body.txt
{
"username":"[email protected]",
"password":"test",
"message":{
"html":"html content goes <b>here</b>",
"text":"text content goes here",
"subject":"this is the subject",
"to":[
{
"email":"[email protected]",
"name":"John Doe"
}
],
"from_email":"[email protected]",
"from_name":"Your Company",
"mailclass":"trans",
"headers":{
"X-foo":"bar"
}
}
}
EOT
curl -X POST -H "Content-Type: application/json" --data-binary @post_body.txt http://127.0.0.1/api/v1/send.json
The API can also be accessed through TLS:
(Remove the --insecure option if you have a valid TLS certificate.)
curl -X POST -H "Content-Type: application/json" --data-binary @post_body.txt --insecure https://127.0.0.1/api/v1/send.json
Compression
The content can be gzip or deflate compressed. This allows payloads with lots of messages to be reduced in size.
This requires the Content-Encoding: gzip or Content-Encoding: deflate header on the HTTP request.
Example:
cat <<'EOT' | gzip -c -9 > post_body.txt.gz
{
"username":"[email protected]",
"password":"test",
"message":{
"html":"html content goes <b>here</b>",
"text":"text content goes here",
"subject":"this is the subject",
"to":[
{
"email":"[email protected]",
"name":"John Doe"
}
],
"from_email":"[email protected]",
"from_name":"Your Company",
"mailclass":"trans",
"headers":{
"X-foo":"bar"
}
}
}
EOT
curl -X POST -H "Content-Type: application/json" -H "Content-Encoding: gzip" --data-binary @post_body.txt.gz http://127.0.0.1/api/v1/send.json
Request Document
|
username
string /required |
Email address of an email user or an Engine User. |
||||||||||||||||||||||||||||||||||||||||||||||||
|
password
string /required |
Password for the user indicated by |
||||||||||||||||||||||||||||||||||||||||||||||||
|
max_request_time
integer /optional |
The maximum number of seconds to spend injecting messages before issuing a response. The actual request time may be slightly longer than this value, as this doesn’t account for network overhead and other latencies in generating the response.
|
||||||||||||||||||||||||||||||||||||||||||||||||
|
message
hash /required This object can either be set directly to the
|
|||||||||||||||||||||||||||||||||||||||||||||||||
Request for Multiple Messages
Up to 500 messages may be submitted per API request.
(Note: Not all messages in a batch are guaranteed to be queued. A response will be provided within max_request_time seconds with as many messages as could be queued in that time. See the section “Message Acceptance Throttling and Back-Pressure” below.)
To submit multiple messages in a single request, provide a document like this:
{
"username":"[email protected]",
"password":"test",
"messages":[
{
"html":"html content goes <b>here</b>",
"text":"text content goes here",
"subject":"this is the subject",
"to":[
{
"email":"[email protected]",
"name":"John Doe"
}
],
"from_email":"[email protected]",
"from_name":"Your Company"
},
{
"html":"html content goes <b>here</b>",
"text":"text content goes here",
"subject":"this is the subject",
"to":[
{
"email":"[email protected]",
"name":"Bob Smith"
}
],
"from_email":"[email protected]",
"from_name":"Your Company"
}
]
}
The JSON document most be an object with the following keys:
| Key | Description |
|---|---|
username |
Same as single message document |
password |
Same as single message document |
messages |
A list containing one or more objects which contain the same information as the message key in a single message document. |
The output can be in two formats.
(1) An error that applies globally to the entire request:
{"success":0,"error":"incorrect username\/password"}
(2) Or the overall request succeeded and a specific success value is provided for each message:
{
"success":1,
"messages":[
{
"success":1,
"message_id":"[email protected]",
"attempted":1,
"id":"1"
},
{
"success":1,
"message_id":"[email protected]",
"attempted":1,
"id":"2"
}
]
}
If there is an internal error, then the subsequent messages in the request will not be attempted. For example:
{
"success":1,
"messages":[
{
"success":1,
"message_id":"[email protected]",
"attempted":1,
"id":"1"
},
{
"success":0,
"error":"internal error: unable to make socket connection (Connection refused)",
"attempted":1,
"id":"2"
},
{
"success":0,
"error":"not attempting due to previous internal errors",
"attempted":0,
"id":"3"
}
]
}
In this request of three messages, the second message encountered an internal error so the third message was not even attempted. The attempted key provides data on if the message was even attempted.
It is very important to handle error conditions and retry messages appropriately.
If the attempted value is 0 for a message, this lets you know that you should retry it.
Here are the two error messages that cause messages to not be attempted:
-
not attempting due to previous internal errors– a previous message encountered an internal error, so the rest of the messages in the batch are not attempted and return this error message. Please retry these messages in a different batch. -
not attempting because previous messages have taken too long– this happens when the system is not able to pre-process messages as fast as you are trying to send them to this API (see “Throttling” below). Only the messages that can be accepted within about 30 seconds will be accepted, and the rest of the messages in the batch will not be attempted and return this error. This is put in place so that an HTTP timeout does not prevent you from getting any answer to the HTTP request.
HTTP keep-alive
For maximum speed, it is recommended to use HTTP keep-alive in your HTTP API, so that you re-use the same TCP connection for each new request.
This can have a significant performance impact because TCP slow-start is bypassed for subsequent requests.
Message Acceptance Throttling and Back-Pressure
Message batches may not be larger than 500 messages.
However, not all messages in a batch are guaranteed to be queued.
The HTTP API will queue as many messages as possible within max_request_time seconds and then provide a status response indicating which messages were queued. (This max_request_time seconds does not include network delay for transferring data.) The purpose of replying within max_request_time seconds is to prevent HTTP timeouts – which would prevent the caller from knowing whether messages were successfully queued or not.
This is how GreenArrow throttles how fast messages are accepted through the HTTP API so that they can not be accepted faster than messages are pre-processed (click/open tracking & DKIM signing) and added to the MTA’s ram-queue. This is a form of “back-pressure,” which is the MTA slowing down the message injection to the speed it can handle.
(Another common form of “back-pressure” is a fraction-of-a-second delay on accepting a message in an SMTP conversation, which has the effect of slowing down the overall speed of messages injection.)
Messages that were unable to be queued due to this limit will be returned with the not attempting because previous messages have taken too long error message (documented above). It is imperative that you re-submit messages returned with this (and the other) error messages documented above.
This error message is the MTA communicating that the batch size is too large.
Recommended batch size
We recommend that your batches be small enough to be accepted within 10 seconds.
There is little benefit to batches that take longer than 10 seconds to be accepted. (At this size, the per-batch overhead is a very small portion of the equation, and larger batches do not gain much if anything.)
There is no benefit (and only a cost) to attempting batches larger than can be handled within max_request_time seconds.
How many messages can be processed within 10 seconds depends on many things, for example: how many other senders are concurrently using the HTTP API, how many messages are currently being injected through other injection methods, and the overall capacity of the system.
We recommend either (a) using a conservative static batch size or (b) dynamically adjusting the batch size based on injection speed.
To dynamically adjust the batch size:
- for the last batch, divide how many messages were accepted by how long the batch took, to calculate the messages per second speed.
- multiply the messages per second speed by 10 to get the size of the next batch
- if larger than 500, reduce to 500
How to observe back-pressure
If message acceptance is currently being throttled due to back-pressure, the size of the Simplemh processing queue in the hvmail_status status output will be above 90%.
Here is an example:
Simplemh processing queue: 98% used ( 1960/ 2000) (messages)
99% used ( 99/ 100) (MB of memory)
Logging API Calls
GreenArrow can write details about HTTP Submission API requests to the Apache error log. To enable writing details to Apache error log, create the file /var/hvmail/control/opt.send_api_debug_log. For example:
echo 1 > /var/hvmail/control/opt.send_api_debug_log
Remove the file to disable logging to Apache’s error log:
rm -f /var/hvmail/control/opt.send_api_debug_log
Apache error log is saved to /var/hvmail/apache/logs/error_log. Here is an example of what GreenArrow logs to Apache’s error log (JOSN objects formatted to improve readability):
[Tue Jun 29 08:08:19.707090 2021] [php7:notice] [pid 19557] [client 1.2.3.4:51269]
http_send_api:
input={
"username": "[email protected]",
"password": "test",
"message": {
"html": "html content goes <b>here</b>",
"text": "text content goes here",
"subject": "this is the subject",
"to": [
{
"email": "[email protected]",
"name": "John Doe"
}
],
"from_email": "[email protected]",
"from_name": "Your Company",
"mailclass": "trans",
"return_path": "[email protected]"
}
}
output={
"success": 0,
"error": "incorrect username/password"
}
-
input=is the JSON object sent to GreenArrow by the API client. -
output=is the JSON object in the response from GreenArrow.