💾 Archived View for jfh.me › posts › 2020-11-28-postfix-postqueue.gmi captured on 2022-04-29 at 12:23:57. Gemini links have been rewritten to link to archived content
⬅️ Previous capture (2021-11-30)
-=-=-=-=-=-=-
I'm starting to learn a bit more about Postfix for work. It's one of those things that I've treated as a black box for a long time. We've been running with basically the same configuration for a very long time. This week we noticed that the active queue backed up with tens of thousands of emails across a few different domains. After waiting for a few hours, we decided to try to purge the emails that were queued up.
By default Postfix has a bunch of commands to give you insight into the queues and where emails are backing up.
https://tecadmin.net/flush-postfix-mail-queue/
http://www.postfix.org/QSHAPE_README.html
I really like the way the `qshape` tool illustrates where the bottleneck in the queue is by pivoting domain and age. My goal was to make a simple script to easily purge emails from any queue based on the age of the email. The easiest starting point is to leverage the output from the `postqueue` tool.
http://www.postfix.org/qshape.1.html
http://www.postfix.org/postqueue.1.html
When you run `postqueue` you get output that looks a bit like this:
-Queue ID- --Size-- ----Arrival Time---- -Sender/Recipient-------
70A5A232A2* 13617 Fri Nov 27 08:34:31 from@foo.com
rcpt@bar.com
731342471C* 11971 Fri Nov 27 10:16:01 from@foo.com
rcpt@bar.com
E3E8A205A9* 42342 Fri Nov 27 16:06:31 from@foo.com
rcpt@bar.com
BC29A24ABB* 25130 Fri Nov 27 14:16:45 from@foo.com
rcpt@bar.com
6072B22E3A* 13541 Fri Nov 27 08:48:14 from@foo.com
rcpt@bar.com
F02C723873* 11961 Fri Nov 27 10:11:32 from@foo.com
rcpt@bar.com
A758123AD9* 11895 Fri Nov 27 10:23:01 from@foo.com
rcpt@bar.com
7A1D324B70* 25182 Fri Nov 27 14:22:44 from@foo.com
rcpt@bar.com
7123D24F59* 13404 Fri Nov 27 10:19:16 from@foo.com
rcpt@bar.com
0140221FE3* 13614 Fri Nov 27 08:32:33 from@foo.com
rcpt@bar.com
930212366F* 13564 Fri Nov 27 11:09:59 from@foo.com
rcpt@bar.com
EBFF5248C1* 25188 Fri Nov 27 14:22:52 from@foo.com
rcpt@bar.com
The output is a little unusual because records are actually separated an empty line. The other thing that's a little unusual about this output is the format of the date. There is no year. My script won't work very well on January first. Right now it's just a maintenance script, but if I wanted to run this all the time in order to regularly purge old emails, I would need to handled that particular situation better.
Here is the initial script.
package main
import (
"bufio"
"flag"
"fmt"
"log"
"os"
"regexp"
"strconv"
"strings"
"time"
)
type (
QueuedItem struct {
QueueID string
Status string
Size int
ArrivalTime time.Time
Sender string
Recipient string
}
)
const (
postfixFormat = "Mon Jan 02 15:04:05"
)
var (
minuteFlag int
)
func main() {
mf := flag.Int("m", 1000, "print a line if the message is older than m minutes")
flag.Parse()
if mf == nil {
log.Fatal("Nil minute flag")
}
minuteFlag = *mf
fromRe := regexp.MustCompile(`^([[:xdigit:]]*)([*!])\s*(\d*)\s*([A-Za-z]*\s[A-Za-z]*\s[0-9]{2}\s[0-9:]*)\s*(.*)
gemini - kennedy.gemi.dev
)
rcptRe := regexp.MustCompile(`^\s*(.*)
gemini - kennedy.gemi.dev
)
scanner := bufio.NewScanner(os.Stdin)
qi := new(QueuedItem)
for scanner.Scan() {
curLine := scanner.Text()
// this is some kind of comme
if strings.HasPrefix(curLine, "-") {
continue
}
// This is a record separator
if curLine == "" {
process(qi)
qi = new(QueuedItem)
}
// this is a from line
fromMatches := fromRe.FindAllStringSubmatch(curLine, -1)
if fromMatches != nil {
var sizeErr error = nil
var timeErr error = nil
qi.QueueID = fromMatches[0][1]
qi.Status = fromMatches[0][2]
qi.Size, sizeErr = strconv.Atoi(fromMatches[0][3])
qi.ArrivalTime, timeErr = time.Parse(postfixFormat, fromMatches[0][4])
qi.Sender = fromMatches[0][5]
if sizeErr != nil {
log.Printf("There was an issue reading the message size: %s", sizeErr.Error())
}
if timeErr != nil {
log.Printf("There was an issue reading the message time: %s", timeErr.Error())
}
continue
}
// this is a to line
rcptMatches := rcptRe.FindAllStringSubmatch(curLine, -1)
if rcptMatches != nil {
qi.Recipient = rcptMatches[0][1]
continue
}
// idk what this is
log.Printf("Unrecognized line: %s", curLine)
}
if err := scanner.Err(); err != nil {
log.Println(err)
}
}
func process(q *QueuedItem) {
// This is is a little weird because postfix dates don't have a year associated with them...
roughNow := time.Now().Format(postfixFormat)
postfixNow, _ := time.Parse(postfixFormat, roughNow)
age := postfixNow.Sub(q.ArrivalTime)
if int(age.Minutes()) < minuteFlag {
// still young
return
}
fmt.Println(q.QueueID)
}
I was considering writing it in perl, but changed my mind because I didn't want to figure out how to parse a date on our production system. I have no idea if any date handling modules or if cpan is available on that system. Go in this instance is helpful because it can be cross compiled and deployed very easily.
In order to purge our queues of old emails I ran something like this:
postqueue -p -c /etc/postfix-02 | ./postqueueparse-linux-amd64 -m 1000 | postsuper -c /etc/postfix-02 -d -
First, I'm getting all of the emails are in the queue for the 02 instance of Postfix. Our system runs multiple instances of Postfix, so the tools need to know which instance I'm working with. I pass that output into the parser script. That will filter down so that it's only outputting the queue IDs of emails that have an arrival time more than 1000 minutes ago. At that point, I can pass the queue IDs into `postsuper -d -` in order to purge the queue and delete any stale emails.