After reconnect, Mosquitto exceeds receive maximum setting sent by broker

[Sarek]

We discovered that when used in an MQTT 5 scenario the mosquitto client library exceeds the receive maximum setting sent by the broker in the CONNACK packet. By default, HiveMQ sends a receive maximum setting of 10 in its answer

Here is some information on how to reproduce with HiveMQ as the broker and using mosquitto_pub to send messages:

1. Create a testfile with 1000 lines

for i in {1..1000}; do echo "line$i" >> /tmp/testfile; done

2. Download and start HiveMQ. Java runtime needs to be at least version 11.

wget https://github.com/hivemq/hivemq-community-edition/releases/download/2020.4/hivemq-ce-2020.4.zip
unzip hivemq-ce-2020.4.zip
cd hivemq-ce-2020.4/bin
./diagnostics.sh

3. In another window, patch and compile mosquitto. The reproduction.patch.txt is just necessary to make mosquitto_pub sleep for 10 seconds, so we have time to stop HiveMQ and force a reconnect.

git clone https://github.com/eclipse/mosquitto
cd mosquitto
git checkout v1.6.12
git am -k ../reconnect.patch.txt
make WITH_WEBSOCKETS=yes DOCDIRS= -j4

4. Start mosquitto_pub and wait for the message Freeze HiveMQ broker within next 10 secs

./client/mosquitto_pub -t mytopic -t clientId1 -q 1 -d -l -D Connect session-expiry-interval 1000000 -V 5 --repeat 10 < /tmp/testfile

5. After the message is displayed, stop the HiveMQ broker

kill -STOP $(pgrep -f hivemq.jar)

6. After the 10 seconds have expired, you should see 10 messages being published. Now kill the HiveMQ broker and start it again:

kill -9 $(pgrep -f hivemq.jar)
./diagnostics.sh

7. mosquitto_pub will now retransmit the original 10 messages, but also continue publishing, until it receives a DISCONNECT packet with reason code 147 ("Receive maximum exceeded").

[ralight]

Thank you for the detailed description and patch, that made it very straightforward. Could you please try the fixes branch, I believe this is now fixed.

[Sarek]

I tested with the fixes branch on a53712a, but I still see the same behaviour that a DISCONNECT with reason code 147 is received.

[Sarek]

Okay, I have to correct this. I mistested and used my system's mosquitto library instead of the newly compiled one. When actually using the new one, it works. Thanks a lot @ralight!

[ralight]

Excellent, thanks for reporting back. I've made that same mistake numerous times.