Skip to main content

headless_lms_server/programs/
email_deliver.rs

1use std::{error::Error as StdError, time::Duration};
2
3use crate::config::program_config::ProgramConfig;
4use crate::prelude::*;
5use anyhow::{Context, Result};
6use chrono::{DateTime, Utc};
7use futures::{FutureExt, StreamExt};
8use headless_lms_models::email_deliveries::{
9    Email, EmailDeliveryErrorInsert, FETCH_LIMIT, RETRY_WINDOW_SECS, fetch_emails,
10    increment_retry_and_mark_non_retryable, increment_retry_and_schedule,
11    insert_email_delivery_error, mark_as_sent, maybe_purge_expired_recipient_addresses,
12};
13use headless_lms_models::email_templates::EmailTemplateType;
14use headless_lms_models::user_email_codes::UserEmailCodePurpose;
15use headless_lms_models::user_passwords::get_unused_reset_password_token_with_user_id;
16use headless_lms_utils::backoff;
17use headless_lms_utils::email_processor::{self, BlockAttributes, EmailGutenbergBlock};
18use lettre::transport::smtp::Error as SmtpError;
19use lettre::transport::smtp::authentication::Credentials;
20use lettre::{
21    Message, SmtpTransport, Transport,
22    message::{MultiPart, SinglePart, header},
23};
24use once_cell::sync::Lazy;
25use secrecy::ExposeSecret;
26use sqlx::{Connection, PgConnection, PgPool};
27use std::collections::HashMap;
28use uuid::Uuid;
29const BATCH_SIZE: usize = FETCH_LIMIT as usize;
30
31/// How often the sender asks the model to purge expired recipient addresses. The model then rolls its
32/// own 1 in 10 chance, so a sweep lands roughly every ten hours.
33const RECIPIENT_ADDRESS_PURGE_INTERVAL: Duration = Duration::from_secs(60 * 60);
34
35const BASE_BACKOFF_SECS: i64 = 60;
36const MAX_BACKOFF_SECS: i64 = 24 * 60 * 60;
37const JITTER_SECS: i64 = 30;
38
39static SMTP_FROM: Lazy<String> = Lazy::new(|| {
40    ProgramConfig::required("SMTP_FROM").expect("No moocfi email found in the env variables.")
41});
42/// The same variable the server reads into `ApplicationConfiguration::base_url`: a mailed link has to
43/// point at the environment that minted its token. Not `FRONTEND_BASE_URL`, which no overlay sets, so
44/// reading that sends dev and test links to production.
45static BASE_URL: Lazy<String> = Lazy::new(|| {
46    ProgramConfig::required("BASE_URL").expect("No BASE_URL found in the env variables.")
47});
48static SMTP_HOST: Lazy<String> = Lazy::new(|| {
49    ProgramConfig::required("SMTP_HOST").expect("No email relay found in the env variables.")
50});
51static DB_URL: Lazy<String> = Lazy::new(|| {
52    ProgramConfig::required("DATABASE_URL").expect("No db url found in the env variables.")
53});
54static SMTP_MESSAGE_ID_DOMAIN: Lazy<String> = Lazy::new(|| {
55    ProgramConfig::optional("SMTP_MESSAGE_ID_DOMAIN")
56        .and_then(|value| {
57            let trimmed = value.trim();
58            if trimmed.is_empty() {
59                None
60            } else {
61                Some(trimmed.to_string())
62            }
63        })
64        .or_else(|| infer_email_domain(SMTP_FROM.as_str()))
65        .unwrap_or_else(|| "courses.mooc.fi".to_string())
66});
67static SMTP_USER: Lazy<String> = Lazy::new(|| {
68    ProgramConfig::required("SMTP_USER").expect("No smtp user found in env variables.")
69});
70static SMTP_PASS: Lazy<String> = Lazy::new(|| {
71    ProgramConfig::required("SMTP_PASS").expect("No smtp password found in env variables.")
72});
73
74pub async fn mail_sender(pool: &PgPool, mailer: &SmtpTransport) -> Result<()> {
75    let mut conn = pool.acquire().await?;
76
77    let emails = fetch_emails(&mut conn).await?;
78
79    let mut futures = tokio_stream::iter(emails)
80        .map(|email| {
81            let email_id = email.id;
82            send_message(email, mailer, pool.clone()).inspect(move |r| {
83                if let Err(err) = r {
84                    tracing::error!("Failed to send email {}: {}", email_id, err)
85                }
86            })
87        })
88        .buffer_unordered(BATCH_SIZE);
89
90    while futures.next().await.is_some() {}
91
92    Ok(())
93}
94
95pub async fn send_message(email: Email, mailer: &SmtpTransport, pool: PgPool) -> Result<()> {
96    let mut conn = pool.acquire().await?;
97    tracing::info!("Email send messages...");
98
99    let now = Utc::now();
100    let attempt = email.retry_count + 1;
101    if retry_window_expired(email.first_failed_at, now) {
102        tracing::warn!(
103            "Retry window expired for email {} (first_failed_at={:?})",
104            email.id,
105            email.first_failed_at
106        );
107        record_non_retryable_failure(
108            &mut conn,
109            email.id,
110            attempt,
111            "retry_window_expired",
112            format!(
113                "Retry window expired before send attempt (email_id={}, user_id={:?}, template={:?}, first_failed_at={:?})",
114                email.id, email.user_id, email.template_type, email.first_failed_at
115            ),
116        )
117        .await?;
118        return Ok(());
119    }
120
121    let mut email_block: Vec<EmailGutenbergBlock> =
122        match email.body.as_ref().context("No body").and_then(|value| {
123            serde_json::from_value(value.clone()).context("Failed to parse email body JSON")
124        }) {
125            Ok(blocks) => blocks,
126            Err(err) => {
127                record_message_build_failure(&mut conn, &email, attempt, &err).await?;
128                return Ok(());
129            }
130        };
131
132    if let Some(template_type) = email.template_type {
133        let template_result = apply_email_template_replacements(
134            &mut conn,
135            template_type,
136            email.id,
137            email.user_id,
138            email.placeholders.as_ref(),
139            email_block,
140            attempt,
141        )
142        .await?;
143        match template_result {
144            TemplateApplyResult::Ready(blocks) => email_block = blocks,
145            TemplateApplyResult::Abandoned => return Ok(()),
146        }
147    }
148
149    let msg_as_plaintext = email_processor::process_content_to_plaintext(&email_block);
150    let msg_as_html = email_processor::process_content_to_html(&email_block);
151
152    let msg = match build_email_message(&email, attempt, msg_as_plaintext, msg_as_html) {
153        Ok(msg) => msg,
154        Err(err) => {
155            record_message_build_failure(&mut conn, &email, attempt, &err).await?;
156            return Ok(());
157        }
158    };
159
160    match mailer.send(&msg) {
161        Ok(_) => {
162            tracing::info!("Email sent successfully {}", email.id);
163            mark_as_sent(&mut conn, email.id)
164                .await
165                .context("Couldn't mark as sent")?;
166        }
167        Err(err) => {
168            let is_transient = is_transient_smtp_error(&err);
169            let (error_code, smtp_response, smtp_response_code) = extract_smtp_error_details(&err);
170
171            tracing::error!(
172                "SMTP send failed for {} (attempt {}, transient={}): {:?}",
173                email.id,
174                attempt,
175                is_transient,
176                err
177            );
178
179            let mut tx = (*conn)
180                .begin()
181                .await
182                .context("Couldn't start email failure transaction")?;
183
184            insert_email_delivery_error(
185                &mut tx,
186                EmailDeliveryErrorInsert {
187                    email_delivery_id: email.id,
188                    attempt,
189                    error_message: err.to_string(),
190                    error_code,
191                    smtp_response,
192                    smtp_response_code,
193                    is_transient,
194                },
195            )
196            .await
197            .context("Couldn't insert email delivery error history")?;
198
199            if is_transient {
200                if retry_window_expired(Some(email.first_failed_at.unwrap_or(now)), now) {
201                    increment_retry_and_mark_non_retryable(&mut tx, email.id)
202                        .await
203                        .context("Couldn't close expired retryable email")?;
204                } else {
205                    // `retry_count` is pre-increment from the claimed row; using it here keeps
206                    // backoff aligned with the next failed-attempt number.
207                    let next_retry_at = compute_next_retry_at(now, email.retry_count);
208                    increment_retry_and_schedule(&mut tx, email.id, Some(next_retry_at))
209                        .await
210                        .context("Couldn't schedule retry")?;
211                }
212            } else {
213                increment_retry_and_mark_non_retryable(&mut tx, email.id)
214                    .await
215                    .context("Couldn't close non-retryable email")?;
216            }
217
218            tx.commit()
219                .await
220                .context("Couldn't commit email failure transaction")?;
221        }
222    };
223
224    Ok(())
225}
226
227enum TemplateApplyResult {
228    Ready(Vec<EmailGutenbergBlock>),
229    Abandoned,
230}
231
232async fn apply_email_template_replacements(
233    conn: &mut PgConnection,
234    template_type: EmailTemplateType,
235    email_id: Uuid,
236    user_id: Option<Uuid>,
237    placeholders: Option<&serde_json::Value>,
238    blocks: Vec<EmailGutenbergBlock>,
239    attempt: i32,
240) -> anyhow::Result<TemplateApplyResult> {
241    let mut replacements = HashMap::new();
242
243    if template_type == EmailTemplateType::Generic {
244        return Ok(TemplateApplyResult::Ready(blocks));
245    }
246
247    if template_type.uses_placeholder_bag() {
248        let replacements = placeholder_bag_replacements(placeholders);
249        return Ok(TemplateApplyResult::Ready(insert_placeholders(
250            blocks,
251            &replacements,
252        )));
253    }
254
255    // Every remaining template derives its values from an account.
256    let Some(user_id) = user_id else {
257        let msg = format!(
258            "Template {template_type:?} requires a user but the delivery is addressed to a raw address"
259        );
260        record_non_retryable_failure(conn, email_id, attempt, "template", msg).await?;
261        return Ok(TemplateApplyResult::Abandoned);
262    };
263
264    match template_type {
265        EmailTemplateType::ResetPasswordEmail => {
266            if let Some(token_str) =
267                get_unused_reset_password_token_with_user_id(conn, user_id).await?
268            {
269                let reset_url = format!(
270                    "{}/reset-user-password/{}",
271                    BASE_URL.trim_end_matches('/'),
272                    token_str.token
273                );
274
275                replacements.insert("RESET_LINK".to_string(), reset_url);
276            } else {
277                let msg = anyhow::anyhow!("No reset token found for user {}", user_id);
278                record_non_retryable_failure(conn, email_id, attempt, "template", msg.to_string())
279                    .await?;
280                return Ok(TemplateApplyResult::Abandoned);
281            }
282        }
283        EmailTemplateType::DeleteUserEmail => {
284            if let Some(code) =
285                headless_lms_models::user_email_codes::get_unused_user_email_code_with_user_id(
286                    conn,
287                    user_id,
288                    UserEmailCodePurpose::AccountDeletion,
289                )
290                .await?
291            {
292                replacements.insert("CODE".to_string(), code.code.expose_secret().to_string());
293            } else {
294                let msg = anyhow::anyhow!("No deletion code found for user {}", user_id);
295                record_non_retryable_failure(conn, email_id, attempt, "template", msg.to_string())
296                    .await?;
297                return Ok(TemplateApplyResult::Abandoned);
298            }
299        }
300        EmailTemplateType::ConfirmEmailCode => {
301            if let Some(code) =
302                headless_lms_models::user_email_codes::get_unused_user_email_code_with_user_id(
303                    conn,
304                    user_id,
305                    UserEmailCodePurpose::AdminLogin,
306                )
307                .await?
308            {
309                replacements.insert("CODE".to_string(), code.code.expose_secret().to_string());
310            } else {
311                let msg = anyhow::anyhow!("No verification code found for user {}", user_id);
312                record_non_retryable_failure(conn, email_id, attempt, "template", msg.to_string())
313                    .await?;
314                return Ok(TemplateApplyResult::Abandoned);
315            }
316        }
317        EmailTemplateType::VerifyEmailAddress => {
318            if let Some(code) =
319                headless_lms_models::user_email_codes::get_unused_user_email_code_with_user_id(
320                    conn,
321                    user_id,
322                    UserEmailCodePurpose::EmailOwnershipVerification,
323                )
324                .await?
325            {
326                replacements.insert("CODE".to_string(), code.code.expose_secret().to_string());
327            } else {
328                let msg = anyhow::anyhow!(
329                    "No email ownership verification code found for user {}",
330                    user_id
331                );
332                record_non_retryable_failure(conn, email_id, attempt, "template", msg.to_string())
333                    .await?;
334                return Ok(TemplateApplyResult::Abandoned);
335            }
336        }
337        // Handled above. Listed rather than caught by `_` so a new template type is a compile error.
338        EmailTemplateType::Generic
339        | EmailTemplateType::CreditRegistrationAccountLinking
340        | EmailTemplateType::CreditRegistrationActionNeeded
341        | EmailTemplateType::CreditRegistrationRegistered
342        | EmailTemplateType::CreditRegistrationStudentNumberLinked => {}
343    }
344
345    Ok(TemplateApplyResult::Ready(insert_placeholders(
346        blocks,
347        &replacements,
348    )))
349}
350
351/// Turns a delivery's placeholder bag into `{{ KEY }}` substitutions. Nested objects and arrays are
352/// skipped: there is no sensible rendering for them in body text.
353fn placeholder_bag_replacements(
354    placeholders: Option<&serde_json::Value>,
355) -> HashMap<String, String> {
356    let Some(serde_json::Value::Object(bag)) = placeholders else {
357        return HashMap::new();
358    };
359    bag.iter()
360        .filter_map(|(key, value)| {
361            let rendered = match value {
362                serde_json::Value::String(s) => s.clone(),
363                serde_json::Value::Number(n) => n.to_string(),
364                serde_json::Value::Bool(b) => b.to_string(),
365                _ => return None,
366            };
367            Some((key.clone(), rendered))
368        })
369        .collect()
370}
371
372fn insert_placeholders(
373    blocks: Vec<EmailGutenbergBlock>,
374    replacements: &HashMap<String, String>,
375) -> Vec<EmailGutenbergBlock> {
376    blocks
377        .into_iter()
378        .map(|mut block| {
379            if let BlockAttributes::Paragraph {
380                content,
381                drop_cap,
382                rest,
383            } = block.attributes
384            {
385                let replaced_content = replacements.iter().fold(content, |acc, (key, value)| {
386                    acc.replace(&format!("{{{{{}}}}}", key), value)
387                });
388
389                block.attributes = BlockAttributes::Paragraph {
390                    content: replaced_content,
391                    drop_cap,
392                    rest,
393                };
394            }
395            block
396        })
397        .collect()
398}
399
400fn build_email_message(
401    email: &Email,
402    attempt: i32,
403    msg_as_plaintext: String,
404    msg_as_html: String,
405) -> Result<Message> {
406    Message::builder()
407        .from(SMTP_FROM.parse()?)
408        .to(email
409            .to
410            .parse()
411            .with_context(|| format!("Invalid recipient address for email_id {}", email.id))?)
412        .subject(email.subject.clone().context("No subject")?)
413        .message_id(Some(format!(
414            "<{}-{}@{}>",
415            email.id,
416            attempt,
417            SMTP_MESSAGE_ID_DOMAIN.as_str()
418        )))
419        .multipart(
420            MultiPart::alternative()
421                .singlepart(
422                    SinglePart::builder()
423                        .header(header::ContentType::TEXT_PLAIN)
424                        .body(msg_as_plaintext),
425                )
426                .singlepart(
427                    SinglePart::builder()
428                        .header(header::ContentType::TEXT_HTML)
429                        .body(msg_as_html),
430                ),
431        )
432        .context("Failed to build email message")
433}
434
435fn infer_email_domain(value: &str) -> Option<String> {
436    let candidate = value
437        .rsplit('<')
438        .next()
439        .unwrap_or(value)
440        .trim()
441        .trim_end_matches('>')
442        .trim();
443    let (_, domain) = candidate.rsplit_once('@')?;
444    let domain = domain.trim();
445    if domain.is_empty() || domain.contains(' ') {
446        None
447    } else {
448        Some(domain.to_string())
449    }
450}
451
452async fn record_message_build_failure(
453    conn: &mut PgConnection,
454    email: &Email,
455    attempt: i32,
456    err: &anyhow::Error,
457) -> Result<()> {
458    tracing::error!(
459        "Message construction failed for email {} (attempt {}): {:#}",
460        email.id,
461        attempt,
462        err
463    );
464    record_non_retryable_failure(
465        conn,
466        email.id,
467        attempt,
468        "message_build",
469        format!("Message construction failed: {err:#}"),
470    )
471    .await
472}
473
474pub async fn main() -> anyhow::Result<()> {
475    tracing_subscriber::fmt().init();
476    dotenvy::dotenv().ok();
477    tracing::info!("Email sender starting up...");
478
479    if ProgramConfig::optional("SMTP_USER").is_none()
480        || ProgramConfig::optional("SMTP_PASS").is_none()
481    {
482        tracing::warn!("SMTP user or password is missing or incorrect");
483    }
484
485    let pool = PgPool::connect(&DB_URL.to_string()).await?;
486    let creds = Credentials::new(SMTP_USER.to_string(), SMTP_PASS.to_string());
487
488    let mailer = match SmtpTransport::relay(&SMTP_HOST) {
489        Ok(builder) => builder.credentials(creds).build(),
490        Err(e) => {
491            tracing::error!("Could not configure SMTP transport: {}", e);
492            return Err(e.into());
493        }
494    };
495
496    let mut interval = tokio::time::interval(Duration::from_secs(10));
497    // Startup counts as an attempt: pods restart often, and firing on the first tick would turn every
498    // restart into another sweep.
499    let mut last_purge_attempt = tokio::time::Instant::now();
500    loop {
501        interval.tick().await;
502        mail_sender(&pool, &mailer).await?;
503
504        // An elapsed check rather than a second interval: another `tick().await` in this loop would
505        // stall the 10 second send cycle until the hour was up and stop mail going out.
506        if last_purge_attempt.elapsed() >= RECIPIENT_ADDRESS_PURGE_INTERVAL {
507            last_purge_attempt = tokio::time::Instant::now();
508            let purged = async {
509                let mut conn = pool.acquire().await?;
510                maybe_purge_expired_recipient_addresses(&mut conn).await
511            }
512            .await;
513            // Logged, not propagated: an error out of main() restarts the pod, which would stop mail
514            // going out over a retention sweep.
515            match purged {
516                Ok(purged) if purged > 0 => {
517                    tracing::info!(
518                        "Purged retained recipient addresses from {purged} email deliveries"
519                    )
520                }
521                Ok(_) => {}
522                Err(err) => tracing::error!("Failed to purge retained recipient addresses: {err}"),
523            }
524        }
525    }
526}
527
528fn retry_window_expired(first_failed_at: Option<DateTime<Utc>>, now: DateTime<Utc>) -> bool {
529    backoff::window_expired(first_failed_at, now, RETRY_WINDOW_SECS)
530}
531
532fn compute_next_retry_at(now: DateTime<Utc>, retry_count: i32) -> DateTime<Utc> {
533    let delay = backoff::exponential_backoff_secs(BASE_BACKOFF_SECS, MAX_BACKOFF_SECS, retry_count);
534    backoff::next_attempt_at(now, delay, JITTER_SECS)
535}
536
537fn is_transient_smtp_error(err: &SmtpError) -> bool {
538    if err.is_transient() {
539        return true;
540    }
541    if err.is_timeout() || err.is_transport_shutdown() {
542        return true;
543    }
544    has_io_error(err)
545}
546
547fn has_io_error(err: &SmtpError) -> bool {
548    let mut source = err.source();
549    while let Some(inner) = source {
550        if inner.is::<std::io::Error>() {
551            return true;
552        }
553        source = inner.source();
554    }
555    false
556}
557
558fn extract_smtp_error_details(err: &SmtpError) -> (Option<String>, Option<String>, Option<i32>) {
559    let smtp_response_code = err.status().map(|code| i32::from(u16::from(code)));
560
561    let error_code = if err.is_transient() {
562        Some("transient".to_string())
563    } else if err.is_permanent() {
564        Some("permanent".to_string())
565    } else if err.is_timeout() {
566        Some("timeout".to_string())
567    } else if has_io_error(err) {
568        Some("network_io".to_string())
569    } else if err.is_transport_shutdown() {
570        Some("transport_shutdown".to_string())
571    } else if err.is_response() {
572        Some("response".to_string())
573    } else if err.is_client() {
574        Some("client".to_string())
575    } else {
576        None
577    };
578
579    let smtp_response = err.source().map(|source| source.to_string());
580
581    (error_code, smtp_response, smtp_response_code)
582}
583
584async fn record_non_retryable_failure(
585    conn: &mut PgConnection,
586    email_id: Uuid,
587    attempt: i32,
588    error_code: &'static str,
589    message: String,
590) -> Result<()> {
591    let mut tx = (*conn)
592        .begin()
593        .await
594        .context("Couldn't start template failure transaction")?;
595
596    insert_email_delivery_error(
597        &mut tx,
598        EmailDeliveryErrorInsert {
599            email_delivery_id: email_id,
600            attempt,
601            error_message: message,
602            error_code: Some(error_code.to_string()),
603            smtp_response: None,
604            smtp_response_code: None,
605            is_transient: false,
606        },
607    )
608    .await
609    .context("Couldn't insert email delivery error history")?;
610
611    increment_retry_and_mark_non_retryable(&mut tx, email_id)
612        .await
613        .context("Couldn't mark email as non-retryable for template error")?;
614
615    tx.commit()
616        .await
617        .context("Couldn't commit template failure transaction")?;
618
619    Ok(())
620}