electrumx memory leak

Another memory leak has been found.

I am using older, stable version on electrumx, since problems with memory leak last time, I settled on older, stable version and stayed there.

commit 835aa492e9a3563d89d99c35168378d65321b849 (HEAD, tag: 1.15.0)
Author: Neil Booth <kyuupichan@gmail.com>
Date:   Wed May 27 11:23:38 2020 +0100

    Prepare 1.15

This memory leak is a lot slower. Process has reached 5 GB RAM after 6 days:

 electrumx_rpc getinfo
{
    "coin": "BitcoinSegwit",
    "daemon": "localhost:8332/",
    "daemon height": 655159,
    "db height": 655159,
    "db_flush_count": 1648,
    "groups": 3115,
    "history cache": "3,243,657 lookups 536,633 hits 1,000 entries",
    "merkle cache": "22,953 lookups 10,313 hits 1,000 entries",
    "peers": {
        "bad": 6,
        "good": 71,
        "never": 12,
        "stale": 14,
        "total": 103
    },
    "pid": 501797,
    "request counts": {
        "blockchain.block.header": 70370,
        "blockchain.block.headers": 2660,
        "blockchain.estimatefee": 23735419,
        "blockchain.headers.subscribe": 135572,
        "blockchain.relayfee": 24953,
        "blockchain.scripthash.get_balance": 283748,
        "blockchain.scripthash.get_history": 155753,
        "blockchain.scripthash.listunspent": 165276,
        "blockchain.scripthash.subscribe": 2920814,
        "blockchain.transaction.broadcast": 1511,
        "blockchain.transaction.get": 43903,
        "blockchain.transaction.get_merkle": 23266,
        "getinfo": 41,
        "mempool.get_fee_histogram": 1897426,
        "server.add_peer": 1146,
        "server.banner": 24969,
        "server.donation_address": 24962,
        "server.features": 10299,
        "server.peers.subscribe": 78146,
        "server.ping": 1464476,
        "server.version": 3696202
    },
    "request total": 34760912,
    "sessions": {
        "count": 931,
        "count with subs": 248,
        "errors": 21,
        "logged": 0,
        "pending requests": 37,
        "subs": 114081
    },
    "tx hashes cache": "23,266 lookups 9,745 hits 1,000 entries",
    "txs sent": 1400,
    "uptime": "6d 12h 41m",
    "version": "ElectrumX 1.15.0"
}

Electrum is running perfectly fine, serving a ton of clients, but memory is slowly creeping, 788 MB per day.

OS: Debian GNU/Linux 10 (buster) x86_64
Kernel: 5.8.0-0.bpo.2-amd64

$ python3 --version
Python 3.7.3

$ pip list |grep aio
aiohttp             3.7.1
aiorpcX             0.18.4

I'd appreciate any help debugging this. What should I upgrade/downgrade first? Waiting for suggestions. I want to make accurate and worthwile test runs, as each test takes few days.

No source file or test is identified. Start by reproducing the leak on ElectrumX 1.15.0 with the reported Python, aiohttp, and aiorpcX versions, recording memory and electrumx_rpc getinfo output over several days. Done means identifying the leaking subsystem, fixing it, and confirming memory remains stable in a comparable long-running test.