Enforce message expiration

[anonimal]

We enforce expiration in some places but we need to review/implement how we enforce, how well we enforce, and if we enforce everything. We should review every layer. Related to #58.

anonimal        zzz: *how* should we enforce message expiration?
+zzz    by dropping it
+zzz    this + bloom filter prevents dups
+zzz    you must add some slop for clock skew, and the amount of slop may depend on where you are (OBEP, IBGW, etc)
+zzz    for example, if at IBEP, it's coming to you, it would be silly to drop aggressively there
+zzz    note that dropping msgs too far in the future is also necessary
fluffypon       makes sense
+zzz    pasta from i2pd-dev:
+zzz    <zzz> ok here's what we do in java:
+zzz    <zzz> not saying it's perfect, and it probably needs to be reviewed.
+zzz    <zzz> I count about a dozen places where we use Router.CLOCK_FUDGE_FACTOR which is defined as 60 seconds
+zzz    <zzz> sometimes we multiply that by 2, or 1.5, or add 3 minutes, or, or, or...
+zzz    <zzz> but +/- 60 seconds seems to be the minimum
+zzz    <zzz> which really isn't a lot, since one router could be +30 seconds off and the other one -30 seconds off.
+zzz    <zzz> like i said, at some places like the IBEP you want more fudge factor, since the message is already where it's going
anonimal        Oh, so we're back to the bloom stuff, ok.
+zzz    right. unless you enforce timestamps, your dup detection is only good for the cycle time of your bloom filter
anonimal        K, another coal for the fire. Seems ticket worthy, will open now. Thank for the details, zzz.
anonimal        zzz: CLOCK_FUDGE_FACTOR is used across several layers/protcols. I haven't look into details, but is our message expiration issue not limited to I2NP?
+zzz    anonimal, yeah there's timestamps or expirations in lots of places, e.g. SSU, leasesets, garlic cloves... don't know where you have issues or where you don't

[anonimal]

NOTICE: THIS ISSUE HAS BEEN MOVED TO GitLab. Please continue the discussion there. See #1013 for details.