That's so cool that you got Fastly to fix this! Was this IPv4 or IPv6?
I stumbled into this exact problem in mid-2019, and my Googling lead to https://tools.ietf.org/html/draft-jaeggli-v6ops-pmtud-ecmp-problem-00, co-authored by someone from Fastly, in which they said that for IPv4, "the relative rarity of sub-1500 byte MTUs [...] makes the problem sufficiently rare that some deployments simply choose to ignore it". After seeing that I assumed we'd have to apply the MSS workaround forever ...
Nice work!
That's so cool that you got Fastly to fix this! Was this IPv4 or IPv6?
I stumbled into this exact problem in mid-2019, and my Googling lead to https://tools.ietf.org/html/draft-jaeggli-v6ops-pmtud-ecmp-problem-00, co-authored by someone from Fastly, in which they said that for IPv4, "the relative rarity of sub-1500 byte MTUs [...] makes the problem sufficiently rare that some deployments simply choose to ignore it". After seeing that I assumed we'd have to apply the MSS workaround forever ...