RabbitMQ memory alarms set and unset until disk space alarm set and unset until crash

We have a RabbitMQ server that runs correctly for a while until it starts setting and unsetting memory alarms back and forth :

380 =INFO REPORT==== 25-Oct-2021::17:46:33 ===
381 vm_memory_high_watermark set. Memory used:3437756080 allowed:3338231808
382 
383 =WARNING REPORT==== 25-Oct-2021::17:46:33 ===
384 memory resource limit alarm set on node 'rabbit@rab-server'.
385 
386 **********************************************************
387 *** Publishers will be blocked until this alarm clears ***
388 **********************************************************
389 
390 =INFO REPORT==== 25-Oct-2021::17:46:36 ===
391 vm_memory_high_watermark clear. Memory used:1541409584 allowed:3338231808
392 
393 =WARNING REPORT==== 25-Oct-2021::17:46:36 ===
394 memory resource limit alarm cleared on node 'rabbit@rab-server'
395 
396 =WARNING REPORT==== 25-Oct-2021::17:46:36 ===
397 memory resource limit alarm cleared across the cluster
398 
399 =INFO REPORT==== 25-Oct-2021::17:46:42 ===
400 vm_memory_high_watermark set. Memory used:4035019336 allowed:3338231808
401 
402 =WARNING REPORT==== 25-Oct-2021::17:46:42 ===
403 memory resource limit alarm set on node 'rabbit@rab-server'.
404 
405 **********************************************************
406 *** Publishers will be blocked until this alarm clears ***
407 **********************************************************
408 
409 =INFO REPORT==== 25-Oct-2021::17:46:45 ===
410 vm_memory_high_watermark clear. Memory used:1786022776 allowed:3338231808

It does so several times and the jumps are always about 2.5GB or memory used.

Until it starts doing the same for disk space

456 =INFO REPORT==== 25-Oct-2021::18:15:35 ===
457 Free disk space is insufficient. Free bytes: 44498944. Limit: 50000000
458 
459 =WARNING REPORT==== 25-Oct-2021::18:15:35 ===
460 disk resource limit alarm set on node 'rabbit@rab-server'.
461 
462 **********************************************************
463 *** Publishers will be blocked until this alarm clears ***
464 **********************************************************
465 
466 =INFO REPORT==== 25-Oct-2021::18:16:05 ===
467 Free disk space is sufficient. Free bytes: 8649433088. Limit: 50000000
468 
469 =WARNING REPORT==== 25-Oct-2021::18:16:05 ===
470 disk resource limit alarm cleared on node 'rabbit@rab-server'
471 
472 =WARNING REPORT==== 25-Oct-2021::18:16:05 ===
473 disk resource limit alarm cleared across the cluster
474 
475 =INFO REPORT==== 25-Oct-2021::18:18:17 ===
476 Free disk space is insufficient. Free bytes: 46092288. Limit: 50000000
477 
478 =WARNING REPORT==== 25-Oct-2021::18:18:17 ===
479 disk resource limit alarm set on node 'rabbit@rab-server'.
480 
481 **********************************************************
482 *** Publishers will be blocked until this alarm clears ***
483 **********************************************************
484 
485 =INFO REPORT==== 25-Oct-2021::18:19:07 ===
486 Free disk space is sufficient. Free bytes: 2646163456. Limit: 50000000
487 
488 =WARNING REPORT==== 25-Oct-2021::18:19:07 ===
489 disk resource limit alarm cleared on node 'rabbit@rab-server'
490 
491 =WARNING REPORT==== 25-Oct-2021::18:19:07 ===
492 disk resource limit alarm cleared across the cluster
493 
494 =INFO REPORT==== 25-Oct-2021::18:40:51 ===
495 Free disk space is insufficient. Free bytes: 49758208. Limit: 50000000
496 
497 =WARNING REPORT==== 25-Oct-2021::18:40:51 ===
498 disk resource limit alarm set on node 'rabbit@rab-server'.
499 
500 **********************************************************
501 *** Publishers will be blocked until this alarm clears ***
502 **********************************************************

And the jump in disk space goes from gigabytes of free disk space to less than 50 megabytes

And then it crashes. Attempting to restart the database in this state will have RabbitMQ try to start and will dump its entire database in the log file, causing the log file to go from about 600 lines to over 19000000 (yes, nineteen millions).

I am having trouble figuring out what is causing this behaviour because when i do

df -h

During the behaviour, none of the disks are full.

I am not an expert on RabbitMQ therefore my questions are :

1 - Does RabbitMQ writes to disk if it has too much in RAM?

2 - Where does RabbitMQ writes? It is possible the 50MB is referring to its writable space?

3 - If you have ever encountered a similar issue, what have you done to fix it?

For now i've added swap space on the machine as it is mentioned to enable it in the memory documentation of RabbitMQ : https://www.rabbitmq.com/memory.html

I had to put the server back on track for now, but i have no way of confirming this will fix the issue until it shows up again.

Thank you for your time.