SSH "kex_exchange_identification: read: Connection reset by peer"

The setup:

• Raspberry 3B running Raspbian Stretch 9 on an external HDD and using ZRAM
• Raspi used as a webserver running LAMP and MERN stacks and accessed remotely via SSH with 1 IDE (Coda for Mac OS)
• SSH port forwarded by router with static IP
• fail2ban running

The problem:

When accessing the raspberry from a remote location (over the Internet) via SSH, it works until the connection hangs. This occurs randomly. I can sometimes SSH it again after few minutes, and sometimes not until I restart the Raspi.

What I've tried:

• SSH in verbose mode from remote location:

debug1: Local version string SSH-2.0-OpenSSH_8.1
kex_exchange_identification: read: Connection reset by peer

• SSH in verbose mode from local network (I actually SSH another machine on the local network remotely, then SSH the Raspi from that machine). Same result: Connection reset by peer
• Checked /etc/hosts.allow and /etc/hosts.deny => Nothing there
• Checked iptables via iptables -L --line-number => Nothing there
• Checked logs: /var/log/fail2ban.log and sudo journalctl -t sshd => Nothing striking there
• Updated sshd_config with no DNS
• Re-installed SSH via apt-get --reinstall install openssh-server openssh-client

I am running out of ideas here and no clue about what's happening. As someone encountered the same problem with SSH connection before ? Could it be a load issue on the raspberry ?

Long story short, my problem had nothing to do with a network issue and was fixed by examining the syslog.

In details:

I noticed that none of the webapps (via LAMP or MERN stacks) while up and running before the issue started, were not reachable anymore.

So I dug up the syslog with the tail -f -n X /var/log/syslog command (replacing X with the number of lines you want to display). I then noticed few lines mentioning a Voltage problem (sorry I did keep the exact terms). But basically it meant that my Raspi which an external HDD was plugged on did not have a strong enough power supply.

Then it looked that the HDD was unmounted and the system crashed, which explains all the issues mentioned above.

So I removed the HDD put the SD card back and ran the Raspi again while going through the syslog again and monitoring the memory with htop. It turned out that when I started both the apache and node servers, the RAM and SWAP memories were getting full repeating the same consequences mentioned above.

So finally I increased the SWAP memory by using ZRAM. Link here .

Now everything runs well but still monitoring.