Mail - Usage & Examples
Practical examples for sending emails and attachments, using templates, queue executors, and verification generators.
Sending emails
Inject IMailService via the MailKeys.MAIL_SERVICE binding key from any service.
Send a single email:
import { BaseService, inject } from '@venizia/ignis';
import { MailKeys, type IMailService } from '@venizia/ignis/mail';
export class UserService extends BaseService {
constructor(
@inject({ key: MailKeys.MAIL_SERVICE })
private _mailService: IMailService,
) {
super({ scope: UserService.name });
}
async sendWelcomeEmail(opts: { userEmail: string; userName: string }) {
const result = await this._mailService.send({
to: opts.userEmail,
subject: 'Welcome to Our App!',
html: `<h1>Welcome ${opts.userName}!</h1><p>Thanks for joining us.</p>`,
text: `Welcome ${opts.userName}! Thanks for joining us.`,
});
if (result.success) {
this.logger.info('[sendWelcomeEmail] Email sent: %s', result.messageId);
} else {
this.logger.error('[sendWelcomeEmail] Failed to send email: %s', result.error);
}
return result;
}
}- Default
from. Ifmessage.fromis omitted,send()fills it fromMailKeys.MAIL_OPTIONS. WithfromNameset, the default renders as"fromName" <from>; with neitherfromnorfromName, it falls back toMailDefaults.FALLBACK_FROM(noreply@example.com). - Transport errors never throw here. The built-in Nodemailer, Mailgun and Amazon SES transports catch their own errors and return
{ success: false, error }.send()only throwsSEND_FAILEDfor a custom transport that throws instead of returning a failed result.
Send a batch of emails:
async sendBulkNotifications(users: Array<{ email: string; name: string }>) {
const messages = users.map(user => ({
to: user.email,
subject: 'Important Update',
html: `<p>Hello ${user.name}, we have an important update for you.</p>`,
}));
const results = await this.mailService.sendBatch(messages, {
concurrency: 5, // Send 5 emails at a time
});
const successCount = results.filter(r => r.success).length;
this.logger.info(
'[sendBulkNotifications] Sent %d/%d emails successfully',
successCount,
results.length,
);
return results;
}- Concurrency defaults to
MailDefaults.BATCH_CONCURRENCY(5).sendBatch()runs every message throughsend()withexecutePromiseWithLimit(). - One bad message never aborts the batch. A
send()that throws is caught per-message and converted to{ success: false, error }; only a failure of the batch operation itself throwsBATCH_SEND_FAILED.
Message validation
MailService.validateMessage() runs before every send. It throws immediately - before the transport is ever called - if any of these hold:
| Condition | Error code | Message |
|---|---|---|
to is missing or an empty array | MailErrorCodes.INVALID_RECIPIENT | Recipient email address is required |
subject is missing | MailErrorCodes.INVALID_CONFIGURATION | Email subject is required |
Both text and html are missing | MailErrorCodes.INVALID_CONFIGURATION | Email must have either text or html content |
text or html is neither a string nor null (a Buffer, a stream, { path }, { href }) | MailErrorCodes.BODY_SOURCE_REFUSED | Mail body refused | text and html must be strings; ... |
// Throws before reaching the transport
await mailService.send({
to: 'user@example.com',
subject: '', // Empty subject triggers validation error
html: '<p>Hello</p>',
});
// Error: { statusCode: 400, message: 'Email subject is required',
// normalized: { code: 'core.mail.invalid_configuration', args: {}, text: 'Email subject is required' } }Attachments
Pass the bytes as content - a Buffer, a Uint8Array, a string, a Node Readable or a web ReadableStream:
await this.mailService.send({
to: 'user@example.com',
subject: 'Your invoice',
html: '<p>Invoice attached.</p>',
attachments: [
{ filename: 'invoice.pdf', content: pdfBuffer, contentType: 'application/pdf' },
{ filename: 'notes.txt', content: 'Paid in full' },
{ filename: 'report.csv', content: createReadStream('./tmp/report.csv') },
],
});To attach a file by path instead, set attachmentRoot on the mail options. A path is then read only from inside that directory:
this.component(MailComponent, {
options: {
provider: MailProviders.NODEMAILER,
config: { host, port, secure, auth },
attachmentRoot: '/srv/app/mail-assets',
},
});
// Reads /srv/app/mail-assets/logo.png; filename and contentType follow from the path
await this.mailService.send({
to: 'user@example.com',
subject: 'Hello',
html: '<img src="cid:logo">',
attachments: [{ path: 'logo.png', cid: 'logo' }],
});MailService.send() turns every attachment into content bytes before the transport sees it, so no transport ever reads a path on its own. The rules:
| Attachment | Result |
|---|---|
path with no attachmentRoot set | Refused - core.mail.attachment_path_refused (400) |
path that resolves outside the root (.., an absolute path elsewhere, a symlink pointing out) | Refused - core.mail.attachment_path_refused (400) |
path to a missing file, a directory, a FIFO or an unreadable file | Refused - the same error and message, so a caller cannot probe which files exist |
path to a file inside the root | Read and sent as content; filename and contentType default from the path |
path as a URL (https://..., data:...) | Treated as a file name under the root, never fetched |
href or raw (Nodemailer sources) | Refused - core.mail.attachment_path_refused (400) |
content that is not bytes, text or a stream | Refused - core.mail.invalid_configuration (400) |
All attachments of one message over maxAttachmentBytes (default 25 MB) | Refused - core.mail.attachment_too_large (413) |
The size limit counts every attachment of the message together, so it also bounds each one. A file is sized before it is opened and counted again while it is read. A stream is counted while it drains: once the running total passes the limit, the read stops and the rest is never buffered. When a send is refused, every attachment stream it was given is destroyed.
The limit counts decoded bytes, and base64 adds about 37% on the wire. The 25 MB default therefore sends about 34 MB, over the 25 MB cap of Gmail and of Mailgun - set maxAttachmentBytes: 18 * 1024 * 1024 for those. Plan memory for about four times the limit per send in flight.
Both options are checked when the component starts. maxAttachmentBytes must be a positive integer, and attachmentRoot an absolute path to an existing directory. Anything else fails the boot with Invalid mail options.
WARNING
Never copy attachments from a request body into a message. Map what the caller uploaded to content yourself. path names a file on your server, so a caller who controls it chooses which file gets mailed. attachmentRoot limits that choice to one directory. It does not make path safe to hand over.
Keep the root writable only by code you trust. Anything that can create links inside it - an archive extraction, a shared upload folder - can swap a directory for a symlink between the check and the read, and a hard link is followed like the file it points to.
The same goes for the body. Nodemailer reads a text or html given as { path } or { href }, so send() accepts only strings there - null counts as absent - and refuses anything else, a Buffer or a stream included, with core.mail.body_source_refused (400). The Nodemailer transport also sets Nodemailer's own disableFileAccess and disableUrlAccess.
Template engine
Register and send a template
Inject both IMailTemplateEngine and IMailService. sendTemplate() renders through the engine and sends via IMailService -- it never bypasses send().
import { BaseService, inject } from '@venizia/ignis';
import { MailKeys, type IMailTemplateEngine, type IMailService } from '@venizia/ignis/mail';
export class NotificationService extends BaseService {
constructor(
@inject({ key: MailKeys.MAIL_TEMPLATE_ENGINE })
private templateEngine: IMailTemplateEngine,
@inject({ key: MailKeys.MAIL_SERVICE })
private mailService: IMailService,
) {
super({ scope: NotificationService.name });
this.registerTemplates();
}
registerTemplates() {
this.templateEngine.registerTemplate({
name: 'welcome-email',
content: `
<html>
<body>
<h1>Welcome {{userName}}!</h1>
<p>Your account has been created successfully.</p>
<p>Your verification code is: <strong>{{verificationCode}}</strong></p>
</body>
</html>
`,
options: {
subject: 'Welcome to {{appName}}',
description: 'Welcome email for new users',
},
});
}
async sendWelcomeEmail(userEmail: string, userName: string, verificationCode: string) {
return this.mailService.sendTemplate({
templateName: 'welcome-email',
data: { userName, verificationCode, appName: 'My Application' },
recipients: userEmail,
options: {
// `path` is read under `attachmentRoot` - see Attachments above
attachments: [{ path: 'logo.png', cid: 'logo' }],
},
});
}
}- Subject resolution order.
options.subjectwins if you pass it. Otherwise the template's ownsubjectwins, rendered through the same engine. If neither is set, the subject falls back to the literal'No Subject'. sendTemplate()requires the template engine binding. It throwsINVALID_CONFIGURATION("Template engine not configured") ifMailKeys.MAIL_TEMPLATE_ENGINEwas never injected. The constructor parameter isisOptional: true, so a service still compiles without it - but the firstsendTemplate()call then fails.
How rendering works
TemplateEngineService keeps templates in an in-memory Map<string, ITemplate>. It substitutes {{variable}} placeholders using the regex /\{\{(\s*[\w.]+\s*)\}\}/g.
- Nested lookup. A key is trimmed, then resolved by splitting on
.and walking the data object (user.profile.name). - Missing values are preserved, not blanked. If a resolved value is
undefinedornull, the original{{placeholder}}text stays in the output, and a warning is logged. The engine never replaces it with an empty string. - String coercion. The engine converts a resolved value with
String(value).
IMPORTANT
Missing template variables are not replaced with empty strings. This makes debugging easier: the rendered output shows you exactly which variables did not resolve.
Validate template data before sending
const template = '<h1>Hello {{userName}}, your code is {{code}}</h1>';
const data = { userName: 'John' }; // Missing 'code'
const validation = this.templateEngine.validateTemplateData({ template, data });
if (!validation.isValid) {
console.error('Missing template variables:', validation.missingKeys); // ['code']
}
// Or throw at render time instead of checking manually
try {
const html = this.templateEngine.render({
templateData: template,
data,
requireValidate: true, // Throws INVALID_CONFIGURATION if any placeholder is missing
});
} catch (error) {
console.error('Template rendering failed:', error.message);
}validateTemplateData() extracts every unique placeholder key from the template and reports:
{
isValid: boolean; // true if all placeholders resolve to a non-null value
missingKeys: string[]; // placeholder names missing from data
allKeys: string[]; // every unique placeholder name found
}Sync templates from a database
async syncTemplatesFromDatabase() {
const templateEngine = this.application.get<IMailTemplateEngine>({
key: MailKeys.MAIL_TEMPLATE_ENGINE,
});
const configRepository = this.application.get<ConfigurationRepository>({
key: 'repositories.ConfigurationRepository',
});
const templateConfigs = await configRepository.find({
filter: { where: { code: { inq: ['MAIL_TEMPLATE_WELCOME', 'MAIL_TEMPLATE_VERIFICATION'] } } },
});
templateConfigs.forEach(config => {
templateEngine.registerTemplate({
name: config.code,
content: config.jValue.content,
options: { subject: config.jValue.subject, description: config.jValue.description },
});
this.logger.info('[syncTemplates] Registered template: %s', config.code);
});
}Queue executors
IMailQueueExecutor is a separate subsystem from MailService. It only exposes enqueueVerificationEmail() and setProcessor(), and it never calls send() on its own. You provide the processor function - typically one that wraps mailService.send(). The executor's job is timing, retry, and delivery guarantees around calling that function.
| Executor | Class | Backing |
|---|---|---|
direct | DirectMailExecutorHelper | None -- calls the processor immediately |
internal-queue | InternalQueueMailExecutorHelper | In-memory SequentialQueueHelper |
bullmq | BullMQMailExecutorHelper | Redis, via BullMQHelper |
Direct executor
The direct executor calls the processor immediately, with no queueing. It returns { queued: false, ... }. If enqueueVerificationEmail() runs before setProcessor(), it throws Processor not set. Call setProcessor() first. Use it for development, or whenever a caller needs a synchronous result.
Internal queue executor
The internal queue executor is in-memory and single-instance, backed by SequentialQueueHelper from @venizia/ignis-helpers with autoDispatch: true.
- Job IDs follow
job_<counter>_<timestamp>. - A
delayoption schedules the enqueue itself viasetTimeout, tracked in adelayedJobsmap. - On failure - a thrown error, or the processor returning
{ success: false }- it retries up tooptions.attempts(default3). - Does not persist jobs across restarts.
close()clears every pending delayed/retry timer.
Retry backoff:
backoff config | Delay |
|---|---|
{ type: 'exponential', delay } | delay * 2^(attempt - 1) |
{ type: 'fixed', delay } | the raw delay |
| Not set | 1000ms |
BullMQ executor
Redis-backed, distributed, backed by BullMQHelper. Job persistence, worker concurrency, prioritization, and delayed execution come from BullMQ itself. removeOnComplete: true, removeOnFail: false (failed jobs stay for debugging). Default enqueue options: attempts: 3, backoff: { type: 'exponential', delay: 1000 }.
Mode gates what the executor can do:
| Mode | Queue created | Workers created | Can enqueue | Can process |
|---|---|---|---|---|
'queue-only' | Yes | No (setProcessor() skips worker creation) | Yes -- without calling setProcessor() first | No |
'worker-only' | No | Yes | No (throws) | Yes |
'both' | Yes | Yes | Yes (requires setProcessor() first) | Yes |
IMPORTANT
'queue-only' mode is the one exception to "call setProcessor() before you enqueue." In that mode, enqueueVerificationEmail() does not need a processor. A producer instance can enqueue jobs that a separate worker-only instance later processes.
Dynamic worker management. Get the bound instance and manage workers at runtime - no restart required:
const executor = this.application.get<BullMQMailExecutorHelper>({
key: MailKeys.MAIL_QUEUE_EXECUTOR_INSTANCE,
});
executor.addWorker({ workerIdentifier: 'mail-queue-worker-extra', concurrency: 10, lockDuration: 60000 });
executor.getWorkerCount(); // e.g. 2
executor.getMode(); // e.g. 'both'
await executor.removeWorker(1); // remove by array index
await executor.clearWorkers(); // close and remove every workersetProcessor() on the BullMQ executor is async and takes an optional second argument for worker configuration. It clears all existing workers before creating new ones:
await executor.setProcessor(
async (email: string) => {
// your processing logic
return { success: true, message: 'Sent', expiresInMinutes: 10 };
},
{
numberOfWorkers: 3, // default: 1
concurrencyPerWorker: 10, // default: 5
lockDuration: 60000, // job lock duration in ms, default: 30000
},
);Verification generators
MailComponent binds three generators. All are transient - a fresh instance per resolution, since none is registered with .setScope('singleton'):
| Generator | Implements | Behavior |
|---|---|---|
NumericCodeGenerator | IVerificationCodeGenerator | crypto.randomInt(0, 10^length), zero-padded via padStart() -- e.g. code 42 at length 6 becomes "000042" |
RandomTokenGenerator | IVerificationTokenGenerator | crypto.randomBytes(bytes).toString('base64url') -- 32 bytes produces a 43-character string, URL-safe, no padding |
DefaultVerificationDataGenerator | IVerificationDataGenerator | Composes both generators via @inject, producing a full IVerificationData with separate expiries |
DefaultVerificationDataGenerator.generateVerificationData() returns:
- A short numeric code (manual entry: SMS, email)
- A long base64url token (URL-based verification links)
- Separate expiries: code via
getExpiryTime(minutes), token viagetExpiryTimeInHours(hours) - ISO 8601 generation timestamps,
codeAttempts: 0,lastCodeSentAtset to now
End-to-end verification flow:
import { BaseService, inject } from '@venizia/ignis';
import { MailKeys, type IMailService, type IVerificationDataGenerator } from '@venizia/ignis/mail';
export class AuthService extends BaseService {
constructor(
@inject({ key: MailKeys.MAIL_SERVICE })
private mailService: IMailService,
@inject({ key: MailKeys.MAIL_VERIFICATION_DATA_GENERATOR })
private verificationGenerator: IVerificationDataGenerator,
) {
super({ scope: AuthService.name });
}
async sendVerificationEmail(userEmail: string) {
const verificationData = this.verificationGenerator.generateVerificationData({
codeLength: 6,
tokenBytes: 32,
codeExpiryMinutes: 10,
tokenExpiryHours: 24,
});
// Persist verificationData to your own user/verification table here.
const result = await this.mailService.send({
to: userEmail,
subject: 'Email Verification',
html: `
<h2>Verify Your Email</h2>
<p>Your verification code is: <strong>${verificationData.verificationCode}</strong></p>
<p>This code expires at: ${verificationData.codeExpiresAt}</p>
<p>Or click this link: https://example.com/verify?token=${verificationData.verificationToken}</p>
`,
});
return { result, verificationData };
}
}Logging and credentials
MailComponent.createAndBindInstances() logs only mailOptions.provider and queueExecutorConfig.type, at info level. It never logs the full config objects, by design. That keeps SMTP passwords, OAuth2 secrets, API keys, and Redis passwords out of the log sink - at least through the component itself.
WARNING
That guarantee only covers what MailComponent logs internally. If your own wrapper component or bootstrap code logs the TMailOptions or IMailQueueExecutorConfig object directly, you reintroduce the leak yourself. This commonly happens while debugging a binding. Log individual safe fields (provider, type) instead of the whole object.
See also
- Setup & Configuration -- quick reference, setup, configuration, and binding keys
- API Reference -- architecture, interfaces, and internals
- Error Reference -- error codes and troubleshooting